1 /*
2  * Copyright 2010 Advanced Micro Devices, Inc.
3  * Copyright 2008 Red Hat Inc.
4  * Copyright 2009 Jerome Glisse.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22  * OTHER DEALINGS IN THE SOFTWARE.
23  *
24  * Authors: Dave Airlie
25  *          Alex Deucher
26  *          Jerome Glisse
27  */
28 #include <drm/drmP.h>
29 #include "radeon.h"
30 #include "radeon_asic.h"
31 #include "evergreend.h"
32 #include "evergreen_reg_safe.h"
33 #include "cayman_reg_safe.h"
34 
35 #define MAX(a,b)                   (((a)>(b))?(a):(b))
36 #define MIN(a,b)                   (((a)<(b))?(a):(b))
37 
38 #define REG_SAFE_BM_SIZE ARRAY_SIZE(evergreen_reg_safe_bm)
39 
40 int r600_dma_cs_next_reloc(struct radeon_cs_parser *p,
41 			   struct radeon_bo_list **cs_reloc);
42 struct evergreen_cs_track {
43 	u32			group_size;
44 	u32			nbanks;
45 	u32			npipes;
46 	u32			row_size;
47 	/* value we track */
48 	u32			nsamples;		/* unused */
49 	struct radeon_bo	*cb_color_bo[12];
50 	u32			cb_color_bo_offset[12];
51 	struct radeon_bo	*cb_color_fmask_bo[8];	/* unused */
52 	struct radeon_bo	*cb_color_cmask_bo[8];	/* unused */
53 	u32			cb_color_info[12];
54 	u32			cb_color_view[12];
55 	u32			cb_color_pitch[12];
56 	u32			cb_color_slice[12];
57 	u32			cb_color_slice_idx[12];
58 	u32			cb_color_attrib[12];
59 	u32			cb_color_cmask_slice[8];/* unused */
60 	u32			cb_color_fmask_slice[8];/* unused */
61 	u32			cb_target_mask;
62 	u32			cb_shader_mask; /* unused */
63 	u32			vgt_strmout_config;
64 	u32			vgt_strmout_buffer_config;
65 	struct radeon_bo	*vgt_strmout_bo[4];
66 	u32			vgt_strmout_bo_offset[4];
67 	u32			vgt_strmout_size[4];
68 	u32			db_depth_control;
69 	u32			db_depth_view;
70 	u32			db_depth_slice;
71 	u32			db_depth_size;
72 	u32			db_z_info;
73 	u32			db_z_read_offset;
74 	u32			db_z_write_offset;
75 	struct radeon_bo	*db_z_read_bo;
76 	struct radeon_bo	*db_z_write_bo;
77 	u32			db_s_info;
78 	u32			db_s_read_offset;
79 	u32			db_s_write_offset;
80 	struct radeon_bo	*db_s_read_bo;
81 	struct radeon_bo	*db_s_write_bo;
82 	bool			sx_misc_kill_all_prims;
83 	bool			cb_dirty;
84 	bool			db_dirty;
85 	bool			streamout_dirty;
86 	u32			htile_offset;
87 	u32			htile_surface;
88 	struct radeon_bo	*htile_bo;
89 	unsigned long		indirect_draw_buffer_size;
90 	const unsigned		*reg_safe_bm;
91 };
92 
93 static u32 evergreen_cs_get_aray_mode(u32 tiling_flags)
94 {
95 	if (tiling_flags & RADEON_TILING_MACRO)
96 		return ARRAY_2D_TILED_THIN1;
97 	else if (tiling_flags & RADEON_TILING_MICRO)
98 		return ARRAY_1D_TILED_THIN1;
99 	else
100 		return ARRAY_LINEAR_GENERAL;
101 }
102 
103 static u32 evergreen_cs_get_num_banks(u32 nbanks)
104 {
105 	switch (nbanks) {
106 	case 2:
107 		return ADDR_SURF_2_BANK;
108 	case 4:
109 		return ADDR_SURF_4_BANK;
110 	case 8:
111 	default:
112 		return ADDR_SURF_8_BANK;
113 	case 16:
114 		return ADDR_SURF_16_BANK;
115 	}
116 }
117 
118 static void evergreen_cs_track_init(struct evergreen_cs_track *track)
119 {
120 	int i;
121 
122 	for (i = 0; i < 8; i++) {
123 		track->cb_color_fmask_bo[i] = NULL;
124 		track->cb_color_cmask_bo[i] = NULL;
125 		track->cb_color_cmask_slice[i] = 0;
126 		track->cb_color_fmask_slice[i] = 0;
127 	}
128 
129 	for (i = 0; i < 12; i++) {
130 		track->cb_color_bo[i] = NULL;
131 		track->cb_color_bo_offset[i] = 0xFFFFFFFF;
132 		track->cb_color_info[i] = 0;
133 		track->cb_color_view[i] = 0xFFFFFFFF;
134 		track->cb_color_pitch[i] = 0;
135 		track->cb_color_slice[i] = 0xfffffff;
136 		track->cb_color_slice_idx[i] = 0;
137 	}
138 	track->cb_target_mask = 0xFFFFFFFF;
139 	track->cb_shader_mask = 0xFFFFFFFF;
140 	track->cb_dirty = true;
141 
142 	track->db_depth_slice = 0xffffffff;
143 	track->db_depth_view = 0xFFFFC000;
144 	track->db_depth_size = 0xFFFFFFFF;
145 	track->db_depth_control = 0xFFFFFFFF;
146 	track->db_z_info = 0xFFFFFFFF;
147 	track->db_z_read_offset = 0xFFFFFFFF;
148 	track->db_z_write_offset = 0xFFFFFFFF;
149 	track->db_z_read_bo = NULL;
150 	track->db_z_write_bo = NULL;
151 	track->db_s_info = 0xFFFFFFFF;
152 	track->db_s_read_offset = 0xFFFFFFFF;
153 	track->db_s_write_offset = 0xFFFFFFFF;
154 	track->db_s_read_bo = NULL;
155 	track->db_s_write_bo = NULL;
156 	track->db_dirty = true;
157 	track->htile_bo = NULL;
158 	track->htile_offset = 0xFFFFFFFF;
159 	track->htile_surface = 0;
160 
161 	for (i = 0; i < 4; i++) {
162 		track->vgt_strmout_size[i] = 0;
163 		track->vgt_strmout_bo[i] = NULL;
164 		track->vgt_strmout_bo_offset[i] = 0xFFFFFFFF;
165 	}
166 	track->streamout_dirty = true;
167 	track->sx_misc_kill_all_prims = false;
168 }
169 
170 struct eg_surface {
171 	/* value gathered from cs */
172 	unsigned	nbx;
173 	unsigned	nby;
174 	unsigned	format;
175 	unsigned	mode;
176 	unsigned	nbanks;
177 	unsigned	bankw;
178 	unsigned	bankh;
179 	unsigned	tsplit;
180 	unsigned	mtilea;
181 	unsigned	nsamples;
182 	/* output value */
183 	unsigned	bpe;
184 	unsigned	layer_size;
185 	unsigned	palign;
186 	unsigned	halign;
187 	unsigned long	base_align;
188 };
189 
190 static int evergreen_surface_check_linear(struct radeon_cs_parser *p,
191 					  struct eg_surface *surf,
192 					  const char *prefix)
193 {
194 	surf->layer_size = surf->nbx * surf->nby * surf->bpe * surf->nsamples;
195 	surf->base_align = surf->bpe;
196 	surf->palign = 1;
197 	surf->halign = 1;
198 	return 0;
199 }
200 
201 static int evergreen_surface_check_linear_aligned(struct radeon_cs_parser *p,
202 						  struct eg_surface *surf,
203 						  const char *prefix)
204 {
205 	struct evergreen_cs_track *track = p->track;
206 	unsigned palign;
207 
208 	palign = MAX(64, track->group_size / surf->bpe);
209 	surf->layer_size = surf->nbx * surf->nby * surf->bpe * surf->nsamples;
210 	surf->base_align = track->group_size;
211 	surf->palign = palign;
212 	surf->halign = 1;
213 	if (surf->nbx & (palign - 1)) {
214 		if (prefix) {
215 			dev_warn(p->dev, "%s:%d %s pitch %d invalid must be aligned with %d\n",
216 				 __func__, __LINE__, prefix, surf->nbx, palign);
217 		}
218 		return -EINVAL;
219 	}
220 	return 0;
221 }
222 
223 static int evergreen_surface_check_1d(struct radeon_cs_parser *p,
224 				      struct eg_surface *surf,
225 				      const char *prefix)
226 {
227 	struct evergreen_cs_track *track = p->track;
228 	unsigned palign;
229 
230 	palign = track->group_size / (8 * surf->bpe * surf->nsamples);
231 	palign = MAX(8, palign);
232 	surf->layer_size = surf->nbx * surf->nby * surf->bpe;
233 	surf->base_align = track->group_size;
234 	surf->palign = palign;
235 	surf->halign = 8;
236 	if ((surf->nbx & (palign - 1))) {
237 		if (prefix) {
238 			dev_warn(p->dev, "%s:%d %s pitch %d invalid must be aligned with %d (%d %d %d)\n",
239 				 __func__, __LINE__, prefix, surf->nbx, palign,
240 				 track->group_size, surf->bpe, surf->nsamples);
241 		}
242 		return -EINVAL;
243 	}
244 	if ((surf->nby & (8 - 1))) {
245 		if (prefix) {
246 			dev_warn(p->dev, "%s:%d %s height %d invalid must be aligned with 8\n",
247 				 __func__, __LINE__, prefix, surf->nby);
248 		}
249 		return -EINVAL;
250 	}
251 	return 0;
252 }
253 
254 static int evergreen_surface_check_2d(struct radeon_cs_parser *p,
255 				      struct eg_surface *surf,
256 				      const char *prefix)
257 {
258 	struct evergreen_cs_track *track = p->track;
259 	unsigned palign, halign, tileb, slice_pt;
260 	unsigned mtile_pr, mtile_ps, mtileb;
261 
262 	tileb = 64 * surf->bpe * surf->nsamples;
263 	slice_pt = 1;
264 	if (tileb > surf->tsplit) {
265 		slice_pt = tileb / surf->tsplit;
266 	}
267 	tileb = tileb / slice_pt;
268 	/* macro tile width & height */
269 	palign = (8 * surf->bankw * track->npipes) * surf->mtilea;
270 	halign = (8 * surf->bankh * surf->nbanks) / surf->mtilea;
271 	mtileb = (palign / 8) * (halign / 8) * tileb;
272 	mtile_pr = surf->nbx / palign;
273 	mtile_ps = (mtile_pr * surf->nby) / halign;
274 	surf->layer_size = mtile_ps * mtileb * slice_pt;
275 	surf->base_align = (palign / 8) * (halign / 8) * tileb;
276 	surf->palign = palign;
277 	surf->halign = halign;
278 
279 	if ((surf->nbx & (palign - 1))) {
280 		if (prefix) {
281 			dev_warn(p->dev, "%s:%d %s pitch %d invalid must be aligned with %d\n",
282 				 __func__, __LINE__, prefix, surf->nbx, palign);
283 		}
284 		return -EINVAL;
285 	}
286 	if ((surf->nby & (halign - 1))) {
287 		if (prefix) {
288 			dev_warn(p->dev, "%s:%d %s height %d invalid must be aligned with %d\n",
289 				 __func__, __LINE__, prefix, surf->nby, halign);
290 		}
291 		return -EINVAL;
292 	}
293 
294 	return 0;
295 }
296 
297 static int evergreen_surface_check(struct radeon_cs_parser *p,
298 				   struct eg_surface *surf,
299 				   const char *prefix)
300 {
301 	/* some common value computed here */
302 	surf->bpe = r600_fmt_get_blocksize(surf->format);
303 
304 	switch (surf->mode) {
305 	case ARRAY_LINEAR_GENERAL:
306 		return evergreen_surface_check_linear(p, surf, prefix);
307 	case ARRAY_LINEAR_ALIGNED:
308 		return evergreen_surface_check_linear_aligned(p, surf, prefix);
309 	case ARRAY_1D_TILED_THIN1:
310 		return evergreen_surface_check_1d(p, surf, prefix);
311 	case ARRAY_2D_TILED_THIN1:
312 		return evergreen_surface_check_2d(p, surf, prefix);
313 	default:
314 		dev_warn(p->dev, "%s:%d %s invalid array mode %d\n",
315 				__func__, __LINE__, prefix, surf->mode);
316 		return -EINVAL;
317 	}
318 	return -EINVAL;
319 }
320 
321 static int evergreen_surface_value_conv_check(struct radeon_cs_parser *p,
322 					      struct eg_surface *surf,
323 					      const char *prefix)
324 {
325 	switch (surf->mode) {
326 	case ARRAY_2D_TILED_THIN1:
327 		break;
328 	case ARRAY_LINEAR_GENERAL:
329 	case ARRAY_LINEAR_ALIGNED:
330 	case ARRAY_1D_TILED_THIN1:
331 		return 0;
332 	default:
333 		dev_warn(p->dev, "%s:%d %s invalid array mode %d\n",
334 				__func__, __LINE__, prefix, surf->mode);
335 		return -EINVAL;
336 	}
337 
338 	switch (surf->nbanks) {
339 	case 0: surf->nbanks = 2; break;
340 	case 1: surf->nbanks = 4; break;
341 	case 2: surf->nbanks = 8; break;
342 	case 3: surf->nbanks = 16; break;
343 	default:
344 		dev_warn(p->dev, "%s:%d %s invalid number of banks %d\n",
345 			 __func__, __LINE__, prefix, surf->nbanks);
346 		return -EINVAL;
347 	}
348 	switch (surf->bankw) {
349 	case 0: surf->bankw = 1; break;
350 	case 1: surf->bankw = 2; break;
351 	case 2: surf->bankw = 4; break;
352 	case 3: surf->bankw = 8; break;
353 	default:
354 		dev_warn(p->dev, "%s:%d %s invalid bankw %d\n",
355 			 __func__, __LINE__, prefix, surf->bankw);
356 		return -EINVAL;
357 	}
358 	switch (surf->bankh) {
359 	case 0: surf->bankh = 1; break;
360 	case 1: surf->bankh = 2; break;
361 	case 2: surf->bankh = 4; break;
362 	case 3: surf->bankh = 8; break;
363 	default:
364 		dev_warn(p->dev, "%s:%d %s invalid bankh %d\n",
365 			 __func__, __LINE__, prefix, surf->bankh);
366 		return -EINVAL;
367 	}
368 	switch (surf->mtilea) {
369 	case 0: surf->mtilea = 1; break;
370 	case 1: surf->mtilea = 2; break;
371 	case 2: surf->mtilea = 4; break;
372 	case 3: surf->mtilea = 8; break;
373 	default:
374 		dev_warn(p->dev, "%s:%d %s invalid macro tile aspect %d\n",
375 			 __func__, __LINE__, prefix, surf->mtilea);
376 		return -EINVAL;
377 	}
378 	switch (surf->tsplit) {
379 	case 0: surf->tsplit = 64; break;
380 	case 1: surf->tsplit = 128; break;
381 	case 2: surf->tsplit = 256; break;
382 	case 3: surf->tsplit = 512; break;
383 	case 4: surf->tsplit = 1024; break;
384 	case 5: surf->tsplit = 2048; break;
385 	case 6: surf->tsplit = 4096; break;
386 	default:
387 		dev_warn(p->dev, "%s:%d %s invalid tile split %d\n",
388 			 __func__, __LINE__, prefix, surf->tsplit);
389 		return -EINVAL;
390 	}
391 	return 0;
392 }
393 
394 static int evergreen_cs_track_validate_cb(struct radeon_cs_parser *p, unsigned id)
395 {
396 	struct evergreen_cs_track *track = p->track;
397 	struct eg_surface surf;
398 	unsigned pitch, slice, mslice;
399 	unsigned long offset;
400 	int r;
401 
402 	mslice = G_028C6C_SLICE_MAX(track->cb_color_view[id]) + 1;
403 	pitch = track->cb_color_pitch[id];
404 	slice = track->cb_color_slice[id];
405 	surf.nbx = (pitch + 1) * 8;
406 	surf.nby = ((slice + 1) * 64) / surf.nbx;
407 	surf.mode = G_028C70_ARRAY_MODE(track->cb_color_info[id]);
408 	surf.format = G_028C70_FORMAT(track->cb_color_info[id]);
409 	surf.tsplit = G_028C74_TILE_SPLIT(track->cb_color_attrib[id]);
410 	surf.nbanks = G_028C74_NUM_BANKS(track->cb_color_attrib[id]);
411 	surf.bankw = G_028C74_BANK_WIDTH(track->cb_color_attrib[id]);
412 	surf.bankh = G_028C74_BANK_HEIGHT(track->cb_color_attrib[id]);
413 	surf.mtilea = G_028C74_MACRO_TILE_ASPECT(track->cb_color_attrib[id]);
414 	surf.nsamples = 1;
415 
416 	if (!r600_fmt_is_valid_color(surf.format)) {
417 		dev_warn(p->dev, "%s:%d cb invalid format %d for %d (0x%08x)\n",
418 			 __func__, __LINE__, surf.format,
419 			id, track->cb_color_info[id]);
420 		return -EINVAL;
421 	}
422 
423 	r = evergreen_surface_value_conv_check(p, &surf, "cb");
424 	if (r) {
425 		return r;
426 	}
427 
428 	r = evergreen_surface_check(p, &surf, "cb");
429 	if (r) {
430 		dev_warn(p->dev, "%s:%d cb[%d] invalid (0x%08x 0x%08x 0x%08x 0x%08x)\n",
431 			 __func__, __LINE__, id, track->cb_color_pitch[id],
432 			 track->cb_color_slice[id], track->cb_color_attrib[id],
433 			 track->cb_color_info[id]);
434 		return r;
435 	}
436 
437 	offset = track->cb_color_bo_offset[id] << 8;
438 	if (offset & (surf.base_align - 1)) {
439 		dev_warn(p->dev, "%s:%d cb[%d] bo base %ld not aligned with %ld\n",
440 			 __func__, __LINE__, id, offset, surf.base_align);
441 		return -EINVAL;
442 	}
443 
444 	offset += surf.layer_size * mslice;
445 	if (offset > radeon_bo_size(track->cb_color_bo[id])) {
446 		/* old ddx are broken they allocate bo with w*h*bpp but
447 		 * program slice with ALIGN(h, 8), catch this and patch
448 		 * command stream.
449 		 */
450 		if (!surf.mode) {
451 			uint32_t *ib = p->ib.ptr;
452 			unsigned long tmp, nby, bsize, size, min = 0;
453 
454 			/* find the height the ddx wants */
455 			if (surf.nby > 8) {
456 				min = surf.nby - 8;
457 			}
458 			bsize = radeon_bo_size(track->cb_color_bo[id]);
459 			tmp = track->cb_color_bo_offset[id] << 8;
460 			for (nby = surf.nby; nby > min; nby--) {
461 				size = nby * surf.nbx * surf.bpe * surf.nsamples;
462 				if ((tmp + size * mslice) <= bsize) {
463 					break;
464 				}
465 			}
466 			if (nby > min) {
467 				surf.nby = nby;
468 				slice = ((nby * surf.nbx) / 64) - 1;
469 				if (!evergreen_surface_check(p, &surf, "cb")) {
470 					/* check if this one works */
471 					tmp += surf.layer_size * mslice;
472 					if (tmp <= bsize) {
473 						ib[track->cb_color_slice_idx[id]] = slice;
474 						goto old_ddx_ok;
475 					}
476 				}
477 			}
478 		}
479 		dev_warn(p->dev, "%s:%d cb[%d] bo too small (layer size %d, "
480 			 "offset %d, max layer %d, bo size %ld, slice %d)\n",
481 			 __func__, __LINE__, id, surf.layer_size,
482 			track->cb_color_bo_offset[id] << 8, mslice,
483 			radeon_bo_size(track->cb_color_bo[id]), slice);
484 		dev_warn(p->dev, "%s:%d problematic surf: (%d %d) (%d %d %d %d %d %d %d)\n",
485 			 __func__, __LINE__, surf.nbx, surf.nby,
486 			surf.mode, surf.bpe, surf.nsamples,
487 			surf.bankw, surf.bankh,
488 			surf.tsplit, surf.mtilea);
489 		return -EINVAL;
490 	}
491 old_ddx_ok:
492 
493 	return 0;
494 }
495 
496 static int evergreen_cs_track_validate_htile(struct radeon_cs_parser *p,
497 						unsigned nbx, unsigned nby)
498 {
499 	struct evergreen_cs_track *track = p->track;
500 	unsigned long size;
501 
502 	if (track->htile_bo == NULL) {
503 		dev_warn(p->dev, "%s:%d htile enabled without htile surface 0x%08x\n",
504 				__func__, __LINE__, track->db_z_info);
505 		return -EINVAL;
506 	}
507 
508 	if (G_028ABC_LINEAR(track->htile_surface)) {
509 		/* pitch must be 16 htiles aligned == 16 * 8 pixel aligned */
510 		nbx = round_up(nbx, 16 * 8);
511 		/* height is npipes htiles aligned == npipes * 8 pixel aligned */
512 		nby = round_up(nby, track->npipes * 8);
513 	} else {
514 		/* always assume 8x8 htile */
515 		/* align is htile align * 8, htile align vary according to
516 		 * number of pipe and tile width and nby
517 		 */
518 		switch (track->npipes) {
519 		case 8:
520 			/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
521 			nbx = round_up(nbx, 64 * 8);
522 			nby = round_up(nby, 64 * 8);
523 			break;
524 		case 4:
525 			/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
526 			nbx = round_up(nbx, 64 * 8);
527 			nby = round_up(nby, 32 * 8);
528 			break;
529 		case 2:
530 			/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
531 			nbx = round_up(nbx, 32 * 8);
532 			nby = round_up(nby, 32 * 8);
533 			break;
534 		case 1:
535 			/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
536 			nbx = round_up(nbx, 32 * 8);
537 			nby = round_up(nby, 16 * 8);
538 			break;
539 		default:
540 			dev_warn(p->dev, "%s:%d invalid num pipes %d\n",
541 					__func__, __LINE__, track->npipes);
542 			return -EINVAL;
543 		}
544 	}
545 	/* compute number of htile */
546 	nbx = nbx >> 3;
547 	nby = nby >> 3;
548 	/* size must be aligned on npipes * 2K boundary */
549 	size = roundup(nbx * nby * 4, track->npipes * (2 << 10));
550 	size += track->htile_offset;
551 
552 	if (size > radeon_bo_size(track->htile_bo)) {
553 		dev_warn(p->dev, "%s:%d htile surface too small %ld for %ld (%d %d)\n",
554 				__func__, __LINE__, radeon_bo_size(track->htile_bo),
555 				size, nbx, nby);
556 		return -EINVAL;
557 	}
558 	return 0;
559 }
560 
561 static int evergreen_cs_track_validate_stencil(struct radeon_cs_parser *p)
562 {
563 	struct evergreen_cs_track *track = p->track;
564 	struct eg_surface surf;
565 	unsigned pitch, slice, mslice;
566 	unsigned long offset;
567 	int r;
568 
569 	mslice = G_028008_SLICE_MAX(track->db_depth_view) + 1;
570 	pitch = G_028058_PITCH_TILE_MAX(track->db_depth_size);
571 	slice = track->db_depth_slice;
572 	surf.nbx = (pitch + 1) * 8;
573 	surf.nby = ((slice + 1) * 64) / surf.nbx;
574 	surf.mode = G_028040_ARRAY_MODE(track->db_z_info);
575 	surf.format = G_028044_FORMAT(track->db_s_info);
576 	surf.tsplit = G_028044_TILE_SPLIT(track->db_s_info);
577 	surf.nbanks = G_028040_NUM_BANKS(track->db_z_info);
578 	surf.bankw = G_028040_BANK_WIDTH(track->db_z_info);
579 	surf.bankh = G_028040_BANK_HEIGHT(track->db_z_info);
580 	surf.mtilea = G_028040_MACRO_TILE_ASPECT(track->db_z_info);
581 	surf.nsamples = 1;
582 
583 	if (surf.format != 1) {
584 		dev_warn(p->dev, "%s:%d stencil invalid format %d\n",
585 			 __func__, __LINE__, surf.format);
586 		return -EINVAL;
587 	}
588 	/* replace by color format so we can use same code */
589 	surf.format = V_028C70_COLOR_8;
590 
591 	r = evergreen_surface_value_conv_check(p, &surf, "stencil");
592 	if (r) {
593 		return r;
594 	}
595 
596 	r = evergreen_surface_check(p, &surf, NULL);
597 	if (r) {
598 		/* old userspace doesn't compute proper depth/stencil alignment
599 		 * check that alignment against a bigger byte per elements and
600 		 * only report if that alignment is wrong too.
601 		 */
602 		surf.format = V_028C70_COLOR_8_8_8_8;
603 		r = evergreen_surface_check(p, &surf, "stencil");
604 		if (r) {
605 			dev_warn(p->dev, "%s:%d stencil invalid (0x%08x 0x%08x 0x%08x 0x%08x)\n",
606 				 __func__, __LINE__, track->db_depth_size,
607 				 track->db_depth_slice, track->db_s_info, track->db_z_info);
608 		}
609 		return r;
610 	}
611 
612 	offset = track->db_s_read_offset << 8;
613 	if (offset & (surf.base_align - 1)) {
614 		dev_warn(p->dev, "%s:%d stencil read bo base %ld not aligned with %ld\n",
615 			 __func__, __LINE__, offset, surf.base_align);
616 		return -EINVAL;
617 	}
618 	offset += surf.layer_size * mslice;
619 	if (offset > radeon_bo_size(track->db_s_read_bo)) {
620 		dev_warn(p->dev, "%s:%d stencil read bo too small (layer size %d, "
621 			 "offset %ld, max layer %d, bo size %ld)\n",
622 			 __func__, __LINE__, surf.layer_size,
623 			(unsigned long)track->db_s_read_offset << 8, mslice,
624 			radeon_bo_size(track->db_s_read_bo));
625 		dev_warn(p->dev, "%s:%d stencil invalid (0x%08x 0x%08x 0x%08x 0x%08x)\n",
626 			 __func__, __LINE__, track->db_depth_size,
627 			 track->db_depth_slice, track->db_s_info, track->db_z_info);
628 		return -EINVAL;
629 	}
630 
631 	offset = track->db_s_write_offset << 8;
632 	if (offset & (surf.base_align - 1)) {
633 		dev_warn(p->dev, "%s:%d stencil write bo base %ld not aligned with %ld\n",
634 			 __func__, __LINE__, offset, surf.base_align);
635 		return -EINVAL;
636 	}
637 	offset += surf.layer_size * mslice;
638 	if (offset > radeon_bo_size(track->db_s_write_bo)) {
639 		dev_warn(p->dev, "%s:%d stencil write bo too small (layer size %d, "
640 			 "offset %ld, max layer %d, bo size %ld)\n",
641 			 __func__, __LINE__, surf.layer_size,
642 			(unsigned long)track->db_s_write_offset << 8, mslice,
643 			radeon_bo_size(track->db_s_write_bo));
644 		return -EINVAL;
645 	}
646 
647 	/* hyperz */
648 	if (G_028040_TILE_SURFACE_ENABLE(track->db_z_info)) {
649 		r = evergreen_cs_track_validate_htile(p, surf.nbx, surf.nby);
650 		if (r) {
651 			return r;
652 		}
653 	}
654 
655 	return 0;
656 }
657 
658 static int evergreen_cs_track_validate_depth(struct radeon_cs_parser *p)
659 {
660 	struct evergreen_cs_track *track = p->track;
661 	struct eg_surface surf;
662 	unsigned pitch, slice, mslice;
663 	unsigned long offset;
664 	int r;
665 
666 	mslice = G_028008_SLICE_MAX(track->db_depth_view) + 1;
667 	pitch = G_028058_PITCH_TILE_MAX(track->db_depth_size);
668 	slice = track->db_depth_slice;
669 	surf.nbx = (pitch + 1) * 8;
670 	surf.nby = ((slice + 1) * 64) / surf.nbx;
671 	surf.mode = G_028040_ARRAY_MODE(track->db_z_info);
672 	surf.format = G_028040_FORMAT(track->db_z_info);
673 	surf.tsplit = G_028040_TILE_SPLIT(track->db_z_info);
674 	surf.nbanks = G_028040_NUM_BANKS(track->db_z_info);
675 	surf.bankw = G_028040_BANK_WIDTH(track->db_z_info);
676 	surf.bankh = G_028040_BANK_HEIGHT(track->db_z_info);
677 	surf.mtilea = G_028040_MACRO_TILE_ASPECT(track->db_z_info);
678 	surf.nsamples = 1;
679 
680 	switch (surf.format) {
681 	case V_028040_Z_16:
682 		surf.format = V_028C70_COLOR_16;
683 		break;
684 	case V_028040_Z_24:
685 	case V_028040_Z_32_FLOAT:
686 		surf.format = V_028C70_COLOR_8_8_8_8;
687 		break;
688 	default:
689 		dev_warn(p->dev, "%s:%d depth invalid format %d\n",
690 			 __func__, __LINE__, surf.format);
691 		return -EINVAL;
692 	}
693 
694 	r = evergreen_surface_value_conv_check(p, &surf, "depth");
695 	if (r) {
696 		dev_warn(p->dev, "%s:%d depth invalid (0x%08x 0x%08x 0x%08x)\n",
697 			 __func__, __LINE__, track->db_depth_size,
698 			 track->db_depth_slice, track->db_z_info);
699 		return r;
700 	}
701 
702 	r = evergreen_surface_check(p, &surf, "depth");
703 	if (r) {
704 		dev_warn(p->dev, "%s:%d depth invalid (0x%08x 0x%08x 0x%08x)\n",
705 			 __func__, __LINE__, track->db_depth_size,
706 			 track->db_depth_slice, track->db_z_info);
707 		return r;
708 	}
709 
710 	offset = track->db_z_read_offset << 8;
711 	if (offset & (surf.base_align - 1)) {
712 		dev_warn(p->dev, "%s:%d stencil read bo base %ld not aligned with %ld\n",
713 			 __func__, __LINE__, offset, surf.base_align);
714 		return -EINVAL;
715 	}
716 	offset += surf.layer_size * mslice;
717 	if (offset > radeon_bo_size(track->db_z_read_bo)) {
718 		dev_warn(p->dev, "%s:%d depth read bo too small (layer size %d, "
719 			 "offset %ld, max layer %d, bo size %ld)\n",
720 			 __func__, __LINE__, surf.layer_size,
721 			(unsigned long)track->db_z_read_offset << 8, mslice,
722 			radeon_bo_size(track->db_z_read_bo));
723 		return -EINVAL;
724 	}
725 
726 	offset = track->db_z_write_offset << 8;
727 	if (offset & (surf.base_align - 1)) {
728 		dev_warn(p->dev, "%s:%d stencil write bo base %ld not aligned with %ld\n",
729 			 __func__, __LINE__, offset, surf.base_align);
730 		return -EINVAL;
731 	}
732 	offset += surf.layer_size * mslice;
733 	if (offset > radeon_bo_size(track->db_z_write_bo)) {
734 		dev_warn(p->dev, "%s:%d depth write bo too small (layer size %d, "
735 			 "offset %ld, max layer %d, bo size %ld)\n",
736 			 __func__, __LINE__, surf.layer_size,
737 			(unsigned long)track->db_z_write_offset << 8, mslice,
738 			radeon_bo_size(track->db_z_write_bo));
739 		return -EINVAL;
740 	}
741 
742 	/* hyperz */
743 	if (G_028040_TILE_SURFACE_ENABLE(track->db_z_info)) {
744 		r = evergreen_cs_track_validate_htile(p, surf.nbx, surf.nby);
745 		if (r) {
746 			return r;
747 		}
748 	}
749 
750 	return 0;
751 }
752 
753 static int evergreen_cs_track_validate_texture(struct radeon_cs_parser *p,
754 					       struct radeon_bo *texture,
755 					       struct radeon_bo *mipmap,
756 					       unsigned idx)
757 {
758 	struct eg_surface surf;
759 	unsigned long toffset, moffset;
760 	unsigned dim, llevel, mslice, width, height, depth, i;
761 	u32 texdw[8];
762 	int r;
763 
764 	texdw[0] = radeon_get_ib_value(p, idx + 0);
765 	texdw[1] = radeon_get_ib_value(p, idx + 1);
766 	texdw[2] = radeon_get_ib_value(p, idx + 2);
767 	texdw[3] = radeon_get_ib_value(p, idx + 3);
768 	texdw[4] = radeon_get_ib_value(p, idx + 4);
769 	texdw[5] = radeon_get_ib_value(p, idx + 5);
770 	texdw[6] = radeon_get_ib_value(p, idx + 6);
771 	texdw[7] = radeon_get_ib_value(p, idx + 7);
772 	dim = G_030000_DIM(texdw[0]);
773 	llevel = G_030014_LAST_LEVEL(texdw[5]);
774 	mslice = G_030014_LAST_ARRAY(texdw[5]) + 1;
775 	width = G_030000_TEX_WIDTH(texdw[0]) + 1;
776 	height =  G_030004_TEX_HEIGHT(texdw[1]) + 1;
777 	depth = G_030004_TEX_DEPTH(texdw[1]) + 1;
778 	surf.format = G_03001C_DATA_FORMAT(texdw[7]);
779 	surf.nbx = (G_030000_PITCH(texdw[0]) + 1) * 8;
780 	surf.nbx = r600_fmt_get_nblocksx(surf.format, surf.nbx);
781 	surf.nby = r600_fmt_get_nblocksy(surf.format, height);
782 	surf.mode = G_030004_ARRAY_MODE(texdw[1]);
783 	surf.tsplit = G_030018_TILE_SPLIT(texdw[6]);
784 	surf.nbanks = G_03001C_NUM_BANKS(texdw[7]);
785 	surf.bankw = G_03001C_BANK_WIDTH(texdw[7]);
786 	surf.bankh = G_03001C_BANK_HEIGHT(texdw[7]);
787 	surf.mtilea = G_03001C_MACRO_TILE_ASPECT(texdw[7]);
788 	surf.nsamples = 1;
789 	toffset = texdw[2] << 8;
790 	moffset = texdw[3] << 8;
791 
792 	if (!r600_fmt_is_valid_texture(surf.format, p->family)) {
793 		dev_warn(p->dev, "%s:%d texture invalid format %d\n",
794 			 __func__, __LINE__, surf.format);
795 		return -EINVAL;
796 	}
797 	switch (dim) {
798 	case V_030000_SQ_TEX_DIM_1D:
799 	case V_030000_SQ_TEX_DIM_2D:
800 	case V_030000_SQ_TEX_DIM_CUBEMAP:
801 	case V_030000_SQ_TEX_DIM_1D_ARRAY:
802 	case V_030000_SQ_TEX_DIM_2D_ARRAY:
803 		depth = 1;
804 		break;
805 	case V_030000_SQ_TEX_DIM_2D_MSAA:
806 	case V_030000_SQ_TEX_DIM_2D_ARRAY_MSAA:
807 		surf.nsamples = 1 << llevel;
808 		llevel = 0;
809 		depth = 1;
810 		break;
811 	case V_030000_SQ_TEX_DIM_3D:
812 		break;
813 	default:
814 		dev_warn(p->dev, "%s:%d texture invalid dimension %d\n",
815 			 __func__, __LINE__, dim);
816 		return -EINVAL;
817 	}
818 
819 	r = evergreen_surface_value_conv_check(p, &surf, "texture");
820 	if (r) {
821 		return r;
822 	}
823 
824 	/* align height */
825 	evergreen_surface_check(p, &surf, NULL);
826 	surf.nby = ALIGN(surf.nby, surf.halign);
827 
828 	r = evergreen_surface_check(p, &surf, "texture");
829 	if (r) {
830 		dev_warn(p->dev, "%s:%d texture invalid 0x%08x 0x%08x 0x%08x 0x%08x 0x%08x 0x%08x\n",
831 			 __func__, __LINE__, texdw[0], texdw[1], texdw[4],
832 			 texdw[5], texdw[6], texdw[7]);
833 		return r;
834 	}
835 
836 	/* check texture size */
837 	if (toffset & (surf.base_align - 1)) {
838 		dev_warn(p->dev, "%s:%d texture bo base %ld not aligned with %ld\n",
839 			 __func__, __LINE__, toffset, surf.base_align);
840 		return -EINVAL;
841 	}
842 	if (surf.nsamples <= 1 && moffset & (surf.base_align - 1)) {
843 		dev_warn(p->dev, "%s:%d mipmap bo base %ld not aligned with %ld\n",
844 			 __func__, __LINE__, moffset, surf.base_align);
845 		return -EINVAL;
846 	}
847 	if (dim == SQ_TEX_DIM_3D) {
848 		toffset += surf.layer_size * depth;
849 	} else {
850 		toffset += surf.layer_size * mslice;
851 	}
852 	if (toffset > radeon_bo_size(texture)) {
853 		dev_warn(p->dev, "%s:%d texture bo too small (layer size %d, "
854 			 "offset %ld, max layer %d, depth %d, bo size %ld) (%d %d)\n",
855 			 __func__, __LINE__, surf.layer_size,
856 			(unsigned long)texdw[2] << 8, mslice,
857 			depth, radeon_bo_size(texture),
858 			surf.nbx, surf.nby);
859 		return -EINVAL;
860 	}
861 
862 	if (!mipmap) {
863 		if (llevel) {
864 			dev_warn(p->dev, "%s:%i got NULL MIP_ADDRESS relocation\n",
865 				 __func__, __LINE__);
866 			return -EINVAL;
867 		} else {
868 			return 0; /* everything's ok */
869 		}
870 	}
871 
872 	/* check mipmap size */
873 	for (i = 1; i <= llevel; i++) {
874 		unsigned w, h, d;
875 
876 		w = r600_mip_minify(width, i);
877 		h = r600_mip_minify(height, i);
878 		d = r600_mip_minify(depth, i);
879 		surf.nbx = r600_fmt_get_nblocksx(surf.format, w);
880 		surf.nby = r600_fmt_get_nblocksy(surf.format, h);
881 
882 		switch (surf.mode) {
883 		case ARRAY_2D_TILED_THIN1:
884 			if (surf.nbx < surf.palign || surf.nby < surf.halign) {
885 				surf.mode = ARRAY_1D_TILED_THIN1;
886 			}
887 			/* recompute alignment */
888 			evergreen_surface_check(p, &surf, NULL);
889 			break;
890 		case ARRAY_LINEAR_GENERAL:
891 		case ARRAY_LINEAR_ALIGNED:
892 		case ARRAY_1D_TILED_THIN1:
893 			break;
894 		default:
895 			dev_warn(p->dev, "%s:%d invalid array mode %d\n",
896 				 __func__, __LINE__, surf.mode);
897 			return -EINVAL;
898 		}
899 		surf.nbx = ALIGN(surf.nbx, surf.palign);
900 		surf.nby = ALIGN(surf.nby, surf.halign);
901 
902 		r = evergreen_surface_check(p, &surf, "mipmap");
903 		if (r) {
904 			return r;
905 		}
906 
907 		if (dim == SQ_TEX_DIM_3D) {
908 			moffset += surf.layer_size * d;
909 		} else {
910 			moffset += surf.layer_size * mslice;
911 		}
912 		if (moffset > radeon_bo_size(mipmap)) {
913 			dev_warn(p->dev, "%s:%d mipmap [%d] bo too small (layer size %d, "
914 					"offset %ld, coffset %ld, max layer %d, depth %d, "
915 					"bo size %ld) level0 (%d %d %d)\n",
916 					__func__, __LINE__, i, surf.layer_size,
917 					(unsigned long)texdw[3] << 8, moffset, mslice,
918 					d, radeon_bo_size(mipmap),
919 					width, height, depth);
920 			dev_warn(p->dev, "%s:%d problematic surf: (%d %d) (%d %d %d %d %d %d %d)\n",
921 				 __func__, __LINE__, surf.nbx, surf.nby,
922 				surf.mode, surf.bpe, surf.nsamples,
923 				surf.bankw, surf.bankh,
924 				surf.tsplit, surf.mtilea);
925 			return -EINVAL;
926 		}
927 	}
928 
929 	return 0;
930 }
931 
932 static int evergreen_cs_track_check(struct radeon_cs_parser *p)
933 {
934 	struct evergreen_cs_track *track = p->track;
935 	unsigned tmp, i;
936 	int r;
937 	unsigned buffer_mask = 0;
938 
939 	/* check streamout */
940 	if (track->streamout_dirty && track->vgt_strmout_config) {
941 		for (i = 0; i < 4; i++) {
942 			if (track->vgt_strmout_config & (1 << i)) {
943 				buffer_mask |= (track->vgt_strmout_buffer_config >> (i * 4)) & 0xf;
944 			}
945 		}
946 
947 		for (i = 0; i < 4; i++) {
948 			if (buffer_mask & (1 << i)) {
949 				if (track->vgt_strmout_bo[i]) {
950 					u64 offset = (u64)track->vgt_strmout_bo_offset[i] +
951 							(u64)track->vgt_strmout_size[i];
952 					if (offset > radeon_bo_size(track->vgt_strmout_bo[i])) {
953 						DRM_ERROR("streamout %d bo too small: 0x%llx, 0x%lx\n",
954 							  i, offset,
955 							  radeon_bo_size(track->vgt_strmout_bo[i]));
956 						return -EINVAL;
957 					}
958 				} else {
959 					dev_warn(p->dev, "No buffer for streamout %d\n", i);
960 					return -EINVAL;
961 				}
962 			}
963 		}
964 		track->streamout_dirty = false;
965 	}
966 
967 	if (track->sx_misc_kill_all_prims)
968 		return 0;
969 
970 	/* check that we have a cb for each enabled target
971 	 */
972 	if (track->cb_dirty) {
973 		tmp = track->cb_target_mask;
974 		for (i = 0; i < 8; i++) {
975 			u32 format = G_028C70_FORMAT(track->cb_color_info[i]);
976 
977 			if (format != V_028C70_COLOR_INVALID &&
978 			    (tmp >> (i * 4)) & 0xF) {
979 				/* at least one component is enabled */
980 				if (track->cb_color_bo[i] == NULL) {
981 					dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
982 						__func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
983 					return -EINVAL;
984 				}
985 				/* check cb */
986 				r = evergreen_cs_track_validate_cb(p, i);
987 				if (r) {
988 					return r;
989 				}
990 			}
991 		}
992 		track->cb_dirty = false;
993 	}
994 
995 	if (track->db_dirty) {
996 		/* Check stencil buffer */
997 		if (G_028044_FORMAT(track->db_s_info) != V_028044_STENCIL_INVALID &&
998 		    G_028800_STENCIL_ENABLE(track->db_depth_control)) {
999 			r = evergreen_cs_track_validate_stencil(p);
1000 			if (r)
1001 				return r;
1002 		}
1003 		/* Check depth buffer */
1004 		if (G_028040_FORMAT(track->db_z_info) != V_028040_Z_INVALID &&
1005 		    G_028800_Z_ENABLE(track->db_depth_control)) {
1006 			r = evergreen_cs_track_validate_depth(p);
1007 			if (r)
1008 				return r;
1009 		}
1010 		track->db_dirty = false;
1011 	}
1012 
1013 	return 0;
1014 }
1015 
1016 /**
1017  * evergreen_cs_packet_parse_vline() - parse userspace VLINE packet
1018  * @parser:		parser structure holding parsing context.
1019  *
1020  * This is an Evergreen(+)-specific function for parsing VLINE packets.
1021  * Real work is done by r600_cs_common_vline_parse function.
1022  * Here we just set up ASIC-specific register table and call
1023  * the common implementation function.
1024  */
1025 static int evergreen_cs_packet_parse_vline(struct radeon_cs_parser *p)
1026 {
1027 
1028 	static uint32_t vline_start_end[6] = {
1029 		EVERGREEN_VLINE_START_END + EVERGREEN_CRTC0_REGISTER_OFFSET,
1030 		EVERGREEN_VLINE_START_END + EVERGREEN_CRTC1_REGISTER_OFFSET,
1031 		EVERGREEN_VLINE_START_END + EVERGREEN_CRTC2_REGISTER_OFFSET,
1032 		EVERGREEN_VLINE_START_END + EVERGREEN_CRTC3_REGISTER_OFFSET,
1033 		EVERGREEN_VLINE_START_END + EVERGREEN_CRTC4_REGISTER_OFFSET,
1034 		EVERGREEN_VLINE_START_END + EVERGREEN_CRTC5_REGISTER_OFFSET
1035 	};
1036 	static uint32_t vline_status[6] = {
1037 		EVERGREEN_VLINE_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET,
1038 		EVERGREEN_VLINE_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET,
1039 		EVERGREEN_VLINE_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET,
1040 		EVERGREEN_VLINE_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET,
1041 		EVERGREEN_VLINE_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET,
1042 		EVERGREEN_VLINE_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET
1043 	};
1044 
1045 	return r600_cs_common_vline_parse(p, vline_start_end, vline_status);
1046 }
1047 
1048 static int evergreen_packet0_check(struct radeon_cs_parser *p,
1049 				   struct radeon_cs_packet *pkt,
1050 				   unsigned idx, unsigned reg)
1051 {
1052 	int r;
1053 
1054 	switch (reg) {
1055 	case EVERGREEN_VLINE_START_END:
1056 		r = evergreen_cs_packet_parse_vline(p);
1057 		if (r) {
1058 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1059 					idx, reg);
1060 			return r;
1061 		}
1062 		break;
1063 	default:
1064 		pr_err("Forbidden register 0x%04X in cs at %d\n", reg, idx);
1065 		return -EINVAL;
1066 	}
1067 	return 0;
1068 }
1069 
1070 static int evergreen_cs_parse_packet0(struct radeon_cs_parser *p,
1071 				      struct radeon_cs_packet *pkt)
1072 {
1073 	unsigned reg, i;
1074 	unsigned idx;
1075 	int r;
1076 
1077 	idx = pkt->idx + 1;
1078 	reg = pkt->reg;
1079 	for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
1080 		r = evergreen_packet0_check(p, pkt, idx, reg);
1081 		if (r) {
1082 			return r;
1083 		}
1084 	}
1085 	return 0;
1086 }
1087 
1088 /**
1089  * evergreen_cs_handle_reg() - process registers that need special handling.
1090  * @parser: parser structure holding parsing context
1091  * @reg: register we are testing
1092  * @idx: index into the cs buffer
1093  */
1094 static int evergreen_cs_handle_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
1095 {
1096 	struct evergreen_cs_track *track = (struct evergreen_cs_track *)p->track;
1097 	struct radeon_bo_list *reloc;
1098 	u32 tmp, *ib;
1099 	int r;
1100 
1101 	ib = p->ib.ptr;
1102 	switch (reg) {
1103 	/* force following reg to 0 in an attempt to disable out buffer
1104 	 * which will need us to better understand how it works to perform
1105 	 * security check on it (Jerome)
1106 	 */
1107 	case SQ_ESGS_RING_SIZE:
1108 	case SQ_GSVS_RING_SIZE:
1109 	case SQ_ESTMP_RING_SIZE:
1110 	case SQ_GSTMP_RING_SIZE:
1111 	case SQ_HSTMP_RING_SIZE:
1112 	case SQ_LSTMP_RING_SIZE:
1113 	case SQ_PSTMP_RING_SIZE:
1114 	case SQ_VSTMP_RING_SIZE:
1115 	case SQ_ESGS_RING_ITEMSIZE:
1116 	case SQ_ESTMP_RING_ITEMSIZE:
1117 	case SQ_GSTMP_RING_ITEMSIZE:
1118 	case SQ_GSVS_RING_ITEMSIZE:
1119 	case SQ_GS_VERT_ITEMSIZE:
1120 	case SQ_GS_VERT_ITEMSIZE_1:
1121 	case SQ_GS_VERT_ITEMSIZE_2:
1122 	case SQ_GS_VERT_ITEMSIZE_3:
1123 	case SQ_GSVS_RING_OFFSET_1:
1124 	case SQ_GSVS_RING_OFFSET_2:
1125 	case SQ_GSVS_RING_OFFSET_3:
1126 	case SQ_HSTMP_RING_ITEMSIZE:
1127 	case SQ_LSTMP_RING_ITEMSIZE:
1128 	case SQ_PSTMP_RING_ITEMSIZE:
1129 	case SQ_VSTMP_RING_ITEMSIZE:
1130 	case VGT_TF_RING_SIZE:
1131 		/* get value to populate the IB don't remove */
1132 		/*tmp =radeon_get_ib_value(p, idx);
1133 		  ib[idx] = 0;*/
1134 		break;
1135 	case SQ_ESGS_RING_BASE:
1136 	case SQ_GSVS_RING_BASE:
1137 	case SQ_ESTMP_RING_BASE:
1138 	case SQ_GSTMP_RING_BASE:
1139 	case SQ_HSTMP_RING_BASE:
1140 	case SQ_LSTMP_RING_BASE:
1141 	case SQ_PSTMP_RING_BASE:
1142 	case SQ_VSTMP_RING_BASE:
1143 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1144 		if (r) {
1145 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1146 					"0x%04X\n", reg);
1147 			return -EINVAL;
1148 		}
1149 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1150 		break;
1151 	case DB_DEPTH_CONTROL:
1152 		track->db_depth_control = radeon_get_ib_value(p, idx);
1153 		track->db_dirty = true;
1154 		break;
1155 	case CAYMAN_DB_EQAA:
1156 		if (p->rdev->family < CHIP_CAYMAN) {
1157 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1158 				 "0x%04X\n", reg);
1159 			return -EINVAL;
1160 		}
1161 		break;
1162 	case CAYMAN_DB_DEPTH_INFO:
1163 		if (p->rdev->family < CHIP_CAYMAN) {
1164 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1165 				 "0x%04X\n", reg);
1166 			return -EINVAL;
1167 		}
1168 		break;
1169 	case DB_Z_INFO:
1170 		track->db_z_info = radeon_get_ib_value(p, idx);
1171 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1172 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1173 			if (r) {
1174 				dev_warn(p->dev, "bad SET_CONTEXT_REG "
1175 						"0x%04X\n", reg);
1176 				return -EINVAL;
1177 			}
1178 			ib[idx] &= ~Z_ARRAY_MODE(0xf);
1179 			track->db_z_info &= ~Z_ARRAY_MODE(0xf);
1180 			ib[idx] |= Z_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
1181 			track->db_z_info |= Z_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
1182 			if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1183 				unsigned bankw, bankh, mtaspect, tile_split;
1184 
1185 				evergreen_tiling_fields(reloc->tiling_flags,
1186 							&bankw, &bankh, &mtaspect,
1187 							&tile_split);
1188 				ib[idx] |= DB_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
1189 				ib[idx] |= DB_TILE_SPLIT(tile_split) |
1190 						DB_BANK_WIDTH(bankw) |
1191 						DB_BANK_HEIGHT(bankh) |
1192 						DB_MACRO_TILE_ASPECT(mtaspect);
1193 			}
1194 		}
1195 		track->db_dirty = true;
1196 		break;
1197 	case DB_STENCIL_INFO:
1198 		track->db_s_info = radeon_get_ib_value(p, idx);
1199 		track->db_dirty = true;
1200 		break;
1201 	case DB_DEPTH_VIEW:
1202 		track->db_depth_view = radeon_get_ib_value(p, idx);
1203 		track->db_dirty = true;
1204 		break;
1205 	case DB_DEPTH_SIZE:
1206 		track->db_depth_size = radeon_get_ib_value(p, idx);
1207 		track->db_dirty = true;
1208 		break;
1209 	case R_02805C_DB_DEPTH_SLICE:
1210 		track->db_depth_slice = radeon_get_ib_value(p, idx);
1211 		track->db_dirty = true;
1212 		break;
1213 	case DB_Z_READ_BASE:
1214 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1215 		if (r) {
1216 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1217 					"0x%04X\n", reg);
1218 			return -EINVAL;
1219 		}
1220 		track->db_z_read_offset = radeon_get_ib_value(p, idx);
1221 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1222 		track->db_z_read_bo = reloc->robj;
1223 		track->db_dirty = true;
1224 		break;
1225 	case DB_Z_WRITE_BASE:
1226 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1227 		if (r) {
1228 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1229 					"0x%04X\n", reg);
1230 			return -EINVAL;
1231 		}
1232 		track->db_z_write_offset = radeon_get_ib_value(p, idx);
1233 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1234 		track->db_z_write_bo = reloc->robj;
1235 		track->db_dirty = true;
1236 		break;
1237 	case DB_STENCIL_READ_BASE:
1238 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1239 		if (r) {
1240 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1241 					"0x%04X\n", reg);
1242 			return -EINVAL;
1243 		}
1244 		track->db_s_read_offset = radeon_get_ib_value(p, idx);
1245 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1246 		track->db_s_read_bo = reloc->robj;
1247 		track->db_dirty = true;
1248 		break;
1249 	case DB_STENCIL_WRITE_BASE:
1250 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1251 		if (r) {
1252 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1253 					"0x%04X\n", reg);
1254 			return -EINVAL;
1255 		}
1256 		track->db_s_write_offset = radeon_get_ib_value(p, idx);
1257 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1258 		track->db_s_write_bo = reloc->robj;
1259 		track->db_dirty = true;
1260 		break;
1261 	case VGT_STRMOUT_CONFIG:
1262 		track->vgt_strmout_config = radeon_get_ib_value(p, idx);
1263 		track->streamout_dirty = true;
1264 		break;
1265 	case VGT_STRMOUT_BUFFER_CONFIG:
1266 		track->vgt_strmout_buffer_config = radeon_get_ib_value(p, idx);
1267 		track->streamout_dirty = true;
1268 		break;
1269 	case VGT_STRMOUT_BUFFER_BASE_0:
1270 	case VGT_STRMOUT_BUFFER_BASE_1:
1271 	case VGT_STRMOUT_BUFFER_BASE_2:
1272 	case VGT_STRMOUT_BUFFER_BASE_3:
1273 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1274 		if (r) {
1275 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1276 					"0x%04X\n", reg);
1277 			return -EINVAL;
1278 		}
1279 		tmp = (reg - VGT_STRMOUT_BUFFER_BASE_0) / 16;
1280 		track->vgt_strmout_bo_offset[tmp] = radeon_get_ib_value(p, idx) << 8;
1281 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1282 		track->vgt_strmout_bo[tmp] = reloc->robj;
1283 		track->streamout_dirty = true;
1284 		break;
1285 	case VGT_STRMOUT_BUFFER_SIZE_0:
1286 	case VGT_STRMOUT_BUFFER_SIZE_1:
1287 	case VGT_STRMOUT_BUFFER_SIZE_2:
1288 	case VGT_STRMOUT_BUFFER_SIZE_3:
1289 		tmp = (reg - VGT_STRMOUT_BUFFER_SIZE_0) / 16;
1290 		/* size in register is DWs, convert to bytes */
1291 		track->vgt_strmout_size[tmp] = radeon_get_ib_value(p, idx) * 4;
1292 		track->streamout_dirty = true;
1293 		break;
1294 	case CP_COHER_BASE:
1295 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1296 		if (r) {
1297 			dev_warn(p->dev, "missing reloc for CP_COHER_BASE "
1298 					"0x%04X\n", reg);
1299 			return -EINVAL;
1300 		}
1301 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1302 	case CB_TARGET_MASK:
1303 		track->cb_target_mask = radeon_get_ib_value(p, idx);
1304 		track->cb_dirty = true;
1305 		break;
1306 	case CB_SHADER_MASK:
1307 		track->cb_shader_mask = radeon_get_ib_value(p, idx);
1308 		track->cb_dirty = true;
1309 		break;
1310 	case PA_SC_AA_CONFIG:
1311 		if (p->rdev->family >= CHIP_CAYMAN) {
1312 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1313 				 "0x%04X\n", reg);
1314 			return -EINVAL;
1315 		}
1316 		tmp = radeon_get_ib_value(p, idx) & MSAA_NUM_SAMPLES_MASK;
1317 		track->nsamples = 1 << tmp;
1318 		break;
1319 	case CAYMAN_PA_SC_AA_CONFIG:
1320 		if (p->rdev->family < CHIP_CAYMAN) {
1321 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1322 				 "0x%04X\n", reg);
1323 			return -EINVAL;
1324 		}
1325 		tmp = radeon_get_ib_value(p, idx) & CAYMAN_MSAA_NUM_SAMPLES_MASK;
1326 		track->nsamples = 1 << tmp;
1327 		break;
1328 	case CB_COLOR0_VIEW:
1329 	case CB_COLOR1_VIEW:
1330 	case CB_COLOR2_VIEW:
1331 	case CB_COLOR3_VIEW:
1332 	case CB_COLOR4_VIEW:
1333 	case CB_COLOR5_VIEW:
1334 	case CB_COLOR6_VIEW:
1335 	case CB_COLOR7_VIEW:
1336 		tmp = (reg - CB_COLOR0_VIEW) / 0x3c;
1337 		track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
1338 		track->cb_dirty = true;
1339 		break;
1340 	case CB_COLOR8_VIEW:
1341 	case CB_COLOR9_VIEW:
1342 	case CB_COLOR10_VIEW:
1343 	case CB_COLOR11_VIEW:
1344 		tmp = ((reg - CB_COLOR8_VIEW) / 0x1c) + 8;
1345 		track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
1346 		track->cb_dirty = true;
1347 		break;
1348 	case CB_COLOR0_INFO:
1349 	case CB_COLOR1_INFO:
1350 	case CB_COLOR2_INFO:
1351 	case CB_COLOR3_INFO:
1352 	case CB_COLOR4_INFO:
1353 	case CB_COLOR5_INFO:
1354 	case CB_COLOR6_INFO:
1355 	case CB_COLOR7_INFO:
1356 		tmp = (reg - CB_COLOR0_INFO) / 0x3c;
1357 		track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1358 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1359 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1360 			if (r) {
1361 				dev_warn(p->dev, "bad SET_CONTEXT_REG "
1362 						"0x%04X\n", reg);
1363 				return -EINVAL;
1364 			}
1365 			ib[idx] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
1366 			track->cb_color_info[tmp] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
1367 		}
1368 		track->cb_dirty = true;
1369 		break;
1370 	case CB_COLOR8_INFO:
1371 	case CB_COLOR9_INFO:
1372 	case CB_COLOR10_INFO:
1373 	case CB_COLOR11_INFO:
1374 		tmp = ((reg - CB_COLOR8_INFO) / 0x1c) + 8;
1375 		track->cb_color_info[tmp] = radeon_get_ib_value(p, idx);
1376 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1377 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1378 			if (r) {
1379 				dev_warn(p->dev, "bad SET_CONTEXT_REG "
1380 						"0x%04X\n", reg);
1381 				return -EINVAL;
1382 			}
1383 			ib[idx] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
1384 			track->cb_color_info[tmp] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
1385 		}
1386 		track->cb_dirty = true;
1387 		break;
1388 	case CB_COLOR0_PITCH:
1389 	case CB_COLOR1_PITCH:
1390 	case CB_COLOR2_PITCH:
1391 	case CB_COLOR3_PITCH:
1392 	case CB_COLOR4_PITCH:
1393 	case CB_COLOR5_PITCH:
1394 	case CB_COLOR6_PITCH:
1395 	case CB_COLOR7_PITCH:
1396 		tmp = (reg - CB_COLOR0_PITCH) / 0x3c;
1397 		track->cb_color_pitch[tmp] = radeon_get_ib_value(p, idx);
1398 		track->cb_dirty = true;
1399 		break;
1400 	case CB_COLOR8_PITCH:
1401 	case CB_COLOR9_PITCH:
1402 	case CB_COLOR10_PITCH:
1403 	case CB_COLOR11_PITCH:
1404 		tmp = ((reg - CB_COLOR8_PITCH) / 0x1c) + 8;
1405 		track->cb_color_pitch[tmp] = radeon_get_ib_value(p, idx);
1406 		track->cb_dirty = true;
1407 		break;
1408 	case CB_COLOR0_SLICE:
1409 	case CB_COLOR1_SLICE:
1410 	case CB_COLOR2_SLICE:
1411 	case CB_COLOR3_SLICE:
1412 	case CB_COLOR4_SLICE:
1413 	case CB_COLOR5_SLICE:
1414 	case CB_COLOR6_SLICE:
1415 	case CB_COLOR7_SLICE:
1416 		tmp = (reg - CB_COLOR0_SLICE) / 0x3c;
1417 		track->cb_color_slice[tmp] = radeon_get_ib_value(p, idx);
1418 		track->cb_color_slice_idx[tmp] = idx;
1419 		track->cb_dirty = true;
1420 		break;
1421 	case CB_COLOR8_SLICE:
1422 	case CB_COLOR9_SLICE:
1423 	case CB_COLOR10_SLICE:
1424 	case CB_COLOR11_SLICE:
1425 		tmp = ((reg - CB_COLOR8_SLICE) / 0x1c) + 8;
1426 		track->cb_color_slice[tmp] = radeon_get_ib_value(p, idx);
1427 		track->cb_color_slice_idx[tmp] = idx;
1428 		track->cb_dirty = true;
1429 		break;
1430 	case CB_COLOR0_ATTRIB:
1431 	case CB_COLOR1_ATTRIB:
1432 	case CB_COLOR2_ATTRIB:
1433 	case CB_COLOR3_ATTRIB:
1434 	case CB_COLOR4_ATTRIB:
1435 	case CB_COLOR5_ATTRIB:
1436 	case CB_COLOR6_ATTRIB:
1437 	case CB_COLOR7_ATTRIB:
1438 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1439 		if (r) {
1440 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1441 					"0x%04X\n", reg);
1442 			return -EINVAL;
1443 		}
1444 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1445 			if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1446 				unsigned bankw, bankh, mtaspect, tile_split;
1447 
1448 				evergreen_tiling_fields(reloc->tiling_flags,
1449 							&bankw, &bankh, &mtaspect,
1450 							&tile_split);
1451 				ib[idx] |= CB_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
1452 				ib[idx] |= CB_TILE_SPLIT(tile_split) |
1453 					   CB_BANK_WIDTH(bankw) |
1454 					   CB_BANK_HEIGHT(bankh) |
1455 					   CB_MACRO_TILE_ASPECT(mtaspect);
1456 			}
1457 		}
1458 		tmp = ((reg - CB_COLOR0_ATTRIB) / 0x3c);
1459 		track->cb_color_attrib[tmp] = ib[idx];
1460 		track->cb_dirty = true;
1461 		break;
1462 	case CB_COLOR8_ATTRIB:
1463 	case CB_COLOR9_ATTRIB:
1464 	case CB_COLOR10_ATTRIB:
1465 	case CB_COLOR11_ATTRIB:
1466 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1467 		if (r) {
1468 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1469 					"0x%04X\n", reg);
1470 			return -EINVAL;
1471 		}
1472 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1473 			if (reloc->tiling_flags & RADEON_TILING_MACRO) {
1474 				unsigned bankw, bankh, mtaspect, tile_split;
1475 
1476 				evergreen_tiling_fields(reloc->tiling_flags,
1477 							&bankw, &bankh, &mtaspect,
1478 							&tile_split);
1479 				ib[idx] |= CB_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
1480 				ib[idx] |= CB_TILE_SPLIT(tile_split) |
1481 					   CB_BANK_WIDTH(bankw) |
1482 					   CB_BANK_HEIGHT(bankh) |
1483 					   CB_MACRO_TILE_ASPECT(mtaspect);
1484 			}
1485 		}
1486 		tmp = ((reg - CB_COLOR8_ATTRIB) / 0x1c) + 8;
1487 		track->cb_color_attrib[tmp] = ib[idx];
1488 		track->cb_dirty = true;
1489 		break;
1490 	case CB_COLOR0_FMASK:
1491 	case CB_COLOR1_FMASK:
1492 	case CB_COLOR2_FMASK:
1493 	case CB_COLOR3_FMASK:
1494 	case CB_COLOR4_FMASK:
1495 	case CB_COLOR5_FMASK:
1496 	case CB_COLOR6_FMASK:
1497 	case CB_COLOR7_FMASK:
1498 		tmp = (reg - CB_COLOR0_FMASK) / 0x3c;
1499 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1500 		if (r) {
1501 			dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1502 			return -EINVAL;
1503 		}
1504 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1505 		track->cb_color_fmask_bo[tmp] = reloc->robj;
1506 		break;
1507 	case CB_COLOR0_CMASK:
1508 	case CB_COLOR1_CMASK:
1509 	case CB_COLOR2_CMASK:
1510 	case CB_COLOR3_CMASK:
1511 	case CB_COLOR4_CMASK:
1512 	case CB_COLOR5_CMASK:
1513 	case CB_COLOR6_CMASK:
1514 	case CB_COLOR7_CMASK:
1515 		tmp = (reg - CB_COLOR0_CMASK) / 0x3c;
1516 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1517 		if (r) {
1518 			dev_err(p->dev, "bad SET_CONTEXT_REG 0x%04X\n", reg);
1519 			return -EINVAL;
1520 		}
1521 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1522 		track->cb_color_cmask_bo[tmp] = reloc->robj;
1523 		break;
1524 	case CB_COLOR0_FMASK_SLICE:
1525 	case CB_COLOR1_FMASK_SLICE:
1526 	case CB_COLOR2_FMASK_SLICE:
1527 	case CB_COLOR3_FMASK_SLICE:
1528 	case CB_COLOR4_FMASK_SLICE:
1529 	case CB_COLOR5_FMASK_SLICE:
1530 	case CB_COLOR6_FMASK_SLICE:
1531 	case CB_COLOR7_FMASK_SLICE:
1532 		tmp = (reg - CB_COLOR0_FMASK_SLICE) / 0x3c;
1533 		track->cb_color_fmask_slice[tmp] = radeon_get_ib_value(p, idx);
1534 		break;
1535 	case CB_COLOR0_CMASK_SLICE:
1536 	case CB_COLOR1_CMASK_SLICE:
1537 	case CB_COLOR2_CMASK_SLICE:
1538 	case CB_COLOR3_CMASK_SLICE:
1539 	case CB_COLOR4_CMASK_SLICE:
1540 	case CB_COLOR5_CMASK_SLICE:
1541 	case CB_COLOR6_CMASK_SLICE:
1542 	case CB_COLOR7_CMASK_SLICE:
1543 		tmp = (reg - CB_COLOR0_CMASK_SLICE) / 0x3c;
1544 		track->cb_color_cmask_slice[tmp] = radeon_get_ib_value(p, idx);
1545 		break;
1546 	case CB_COLOR0_BASE:
1547 	case CB_COLOR1_BASE:
1548 	case CB_COLOR2_BASE:
1549 	case CB_COLOR3_BASE:
1550 	case CB_COLOR4_BASE:
1551 	case CB_COLOR5_BASE:
1552 	case CB_COLOR6_BASE:
1553 	case CB_COLOR7_BASE:
1554 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1555 		if (r) {
1556 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1557 					"0x%04X\n", reg);
1558 			return -EINVAL;
1559 		}
1560 		tmp = (reg - CB_COLOR0_BASE) / 0x3c;
1561 		track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx);
1562 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1563 		track->cb_color_bo[tmp] = reloc->robj;
1564 		track->cb_dirty = true;
1565 		break;
1566 	case CB_COLOR8_BASE:
1567 	case CB_COLOR9_BASE:
1568 	case CB_COLOR10_BASE:
1569 	case CB_COLOR11_BASE:
1570 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1571 		if (r) {
1572 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1573 					"0x%04X\n", reg);
1574 			return -EINVAL;
1575 		}
1576 		tmp = ((reg - CB_COLOR8_BASE) / 0x1c) + 8;
1577 		track->cb_color_bo_offset[tmp] = radeon_get_ib_value(p, idx);
1578 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1579 		track->cb_color_bo[tmp] = reloc->robj;
1580 		track->cb_dirty = true;
1581 		break;
1582 	case DB_HTILE_DATA_BASE:
1583 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1584 		if (r) {
1585 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1586 					"0x%04X\n", reg);
1587 			return -EINVAL;
1588 		}
1589 		track->htile_offset = radeon_get_ib_value(p, idx);
1590 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1591 		track->htile_bo = reloc->robj;
1592 		track->db_dirty = true;
1593 		break;
1594 	case DB_HTILE_SURFACE:
1595 		/* 8x8 only */
1596 		track->htile_surface = radeon_get_ib_value(p, idx);
1597 		/* force 8x8 htile width and height */
1598 		ib[idx] |= 3;
1599 		track->db_dirty = true;
1600 		break;
1601 	case CB_IMMED0_BASE:
1602 	case CB_IMMED1_BASE:
1603 	case CB_IMMED2_BASE:
1604 	case CB_IMMED3_BASE:
1605 	case CB_IMMED4_BASE:
1606 	case CB_IMMED5_BASE:
1607 	case CB_IMMED6_BASE:
1608 	case CB_IMMED7_BASE:
1609 	case CB_IMMED8_BASE:
1610 	case CB_IMMED9_BASE:
1611 	case CB_IMMED10_BASE:
1612 	case CB_IMMED11_BASE:
1613 	case SQ_PGM_START_FS:
1614 	case SQ_PGM_START_ES:
1615 	case SQ_PGM_START_VS:
1616 	case SQ_PGM_START_GS:
1617 	case SQ_PGM_START_PS:
1618 	case SQ_PGM_START_HS:
1619 	case SQ_PGM_START_LS:
1620 	case SQ_CONST_MEM_BASE:
1621 	case SQ_ALU_CONST_CACHE_GS_0:
1622 	case SQ_ALU_CONST_CACHE_GS_1:
1623 	case SQ_ALU_CONST_CACHE_GS_2:
1624 	case SQ_ALU_CONST_CACHE_GS_3:
1625 	case SQ_ALU_CONST_CACHE_GS_4:
1626 	case SQ_ALU_CONST_CACHE_GS_5:
1627 	case SQ_ALU_CONST_CACHE_GS_6:
1628 	case SQ_ALU_CONST_CACHE_GS_7:
1629 	case SQ_ALU_CONST_CACHE_GS_8:
1630 	case SQ_ALU_CONST_CACHE_GS_9:
1631 	case SQ_ALU_CONST_CACHE_GS_10:
1632 	case SQ_ALU_CONST_CACHE_GS_11:
1633 	case SQ_ALU_CONST_CACHE_GS_12:
1634 	case SQ_ALU_CONST_CACHE_GS_13:
1635 	case SQ_ALU_CONST_CACHE_GS_14:
1636 	case SQ_ALU_CONST_CACHE_GS_15:
1637 	case SQ_ALU_CONST_CACHE_PS_0:
1638 	case SQ_ALU_CONST_CACHE_PS_1:
1639 	case SQ_ALU_CONST_CACHE_PS_2:
1640 	case SQ_ALU_CONST_CACHE_PS_3:
1641 	case SQ_ALU_CONST_CACHE_PS_4:
1642 	case SQ_ALU_CONST_CACHE_PS_5:
1643 	case SQ_ALU_CONST_CACHE_PS_6:
1644 	case SQ_ALU_CONST_CACHE_PS_7:
1645 	case SQ_ALU_CONST_CACHE_PS_8:
1646 	case SQ_ALU_CONST_CACHE_PS_9:
1647 	case SQ_ALU_CONST_CACHE_PS_10:
1648 	case SQ_ALU_CONST_CACHE_PS_11:
1649 	case SQ_ALU_CONST_CACHE_PS_12:
1650 	case SQ_ALU_CONST_CACHE_PS_13:
1651 	case SQ_ALU_CONST_CACHE_PS_14:
1652 	case SQ_ALU_CONST_CACHE_PS_15:
1653 	case SQ_ALU_CONST_CACHE_VS_0:
1654 	case SQ_ALU_CONST_CACHE_VS_1:
1655 	case SQ_ALU_CONST_CACHE_VS_2:
1656 	case SQ_ALU_CONST_CACHE_VS_3:
1657 	case SQ_ALU_CONST_CACHE_VS_4:
1658 	case SQ_ALU_CONST_CACHE_VS_5:
1659 	case SQ_ALU_CONST_CACHE_VS_6:
1660 	case SQ_ALU_CONST_CACHE_VS_7:
1661 	case SQ_ALU_CONST_CACHE_VS_8:
1662 	case SQ_ALU_CONST_CACHE_VS_9:
1663 	case SQ_ALU_CONST_CACHE_VS_10:
1664 	case SQ_ALU_CONST_CACHE_VS_11:
1665 	case SQ_ALU_CONST_CACHE_VS_12:
1666 	case SQ_ALU_CONST_CACHE_VS_13:
1667 	case SQ_ALU_CONST_CACHE_VS_14:
1668 	case SQ_ALU_CONST_CACHE_VS_15:
1669 	case SQ_ALU_CONST_CACHE_HS_0:
1670 	case SQ_ALU_CONST_CACHE_HS_1:
1671 	case SQ_ALU_CONST_CACHE_HS_2:
1672 	case SQ_ALU_CONST_CACHE_HS_3:
1673 	case SQ_ALU_CONST_CACHE_HS_4:
1674 	case SQ_ALU_CONST_CACHE_HS_5:
1675 	case SQ_ALU_CONST_CACHE_HS_6:
1676 	case SQ_ALU_CONST_CACHE_HS_7:
1677 	case SQ_ALU_CONST_CACHE_HS_8:
1678 	case SQ_ALU_CONST_CACHE_HS_9:
1679 	case SQ_ALU_CONST_CACHE_HS_10:
1680 	case SQ_ALU_CONST_CACHE_HS_11:
1681 	case SQ_ALU_CONST_CACHE_HS_12:
1682 	case SQ_ALU_CONST_CACHE_HS_13:
1683 	case SQ_ALU_CONST_CACHE_HS_14:
1684 	case SQ_ALU_CONST_CACHE_HS_15:
1685 	case SQ_ALU_CONST_CACHE_LS_0:
1686 	case SQ_ALU_CONST_CACHE_LS_1:
1687 	case SQ_ALU_CONST_CACHE_LS_2:
1688 	case SQ_ALU_CONST_CACHE_LS_3:
1689 	case SQ_ALU_CONST_CACHE_LS_4:
1690 	case SQ_ALU_CONST_CACHE_LS_5:
1691 	case SQ_ALU_CONST_CACHE_LS_6:
1692 	case SQ_ALU_CONST_CACHE_LS_7:
1693 	case SQ_ALU_CONST_CACHE_LS_8:
1694 	case SQ_ALU_CONST_CACHE_LS_9:
1695 	case SQ_ALU_CONST_CACHE_LS_10:
1696 	case SQ_ALU_CONST_CACHE_LS_11:
1697 	case SQ_ALU_CONST_CACHE_LS_12:
1698 	case SQ_ALU_CONST_CACHE_LS_13:
1699 	case SQ_ALU_CONST_CACHE_LS_14:
1700 	case SQ_ALU_CONST_CACHE_LS_15:
1701 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1702 		if (r) {
1703 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1704 					"0x%04X\n", reg);
1705 			return -EINVAL;
1706 		}
1707 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1708 		break;
1709 	case SX_MEMORY_EXPORT_BASE:
1710 		if (p->rdev->family >= CHIP_CAYMAN) {
1711 			dev_warn(p->dev, "bad SET_CONFIG_REG "
1712 				 "0x%04X\n", reg);
1713 			return -EINVAL;
1714 		}
1715 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1716 		if (r) {
1717 			dev_warn(p->dev, "bad SET_CONFIG_REG "
1718 					"0x%04X\n", reg);
1719 			return -EINVAL;
1720 		}
1721 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1722 		break;
1723 	case CAYMAN_SX_SCATTER_EXPORT_BASE:
1724 		if (p->rdev->family < CHIP_CAYMAN) {
1725 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1726 				 "0x%04X\n", reg);
1727 			return -EINVAL;
1728 		}
1729 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1730 		if (r) {
1731 			dev_warn(p->dev, "bad SET_CONTEXT_REG "
1732 					"0x%04X\n", reg);
1733 			return -EINVAL;
1734 		}
1735 		ib[idx] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
1736 		break;
1737 	case SX_MISC:
1738 		track->sx_misc_kill_all_prims = (radeon_get_ib_value(p, idx) & 0x1) != 0;
1739 		break;
1740 	default:
1741 		dev_warn(p->dev, "forbidden register 0x%08x at %d\n", reg, idx);
1742 		return -EINVAL;
1743 	}
1744 	return 0;
1745 }
1746 
1747 /**
1748  * evergreen_is_safe_reg() - check if register is authorized or not
1749  * @parser: parser structure holding parsing context
1750  * @reg: register we are testing
1751  *
1752  * This function will test against reg_safe_bm and return true
1753  * if register is safe or false otherwise.
1754  */
1755 static inline bool evergreen_is_safe_reg(struct radeon_cs_parser *p, u32 reg)
1756 {
1757 	struct evergreen_cs_track *track = p->track;
1758 	u32 m, i;
1759 
1760 	i = (reg >> 7);
1761 	if (unlikely(i >= REG_SAFE_BM_SIZE)) {
1762 		return false;
1763 	}
1764 	m = 1 << ((reg >> 2) & 31);
1765 	if (!(track->reg_safe_bm[i] & m))
1766 		return true;
1767 
1768 	return false;
1769 }
1770 
1771 static int evergreen_packet3_check(struct radeon_cs_parser *p,
1772 				   struct radeon_cs_packet *pkt)
1773 {
1774 	struct radeon_bo_list *reloc;
1775 	struct evergreen_cs_track *track;
1776 	uint32_t *ib;
1777 	unsigned idx;
1778 	unsigned i;
1779 	unsigned start_reg, end_reg, reg;
1780 	int r;
1781 	u32 idx_value;
1782 
1783 	track = (struct evergreen_cs_track *)p->track;
1784 	ib = p->ib.ptr;
1785 	idx = pkt->idx + 1;
1786 	idx_value = radeon_get_ib_value(p, idx);
1787 
1788 	switch (pkt->opcode) {
1789 	case PACKET3_SET_PREDICATION:
1790 	{
1791 		int pred_op;
1792 		int tmp;
1793 		uint64_t offset;
1794 
1795 		if (pkt->count != 1) {
1796 			DRM_ERROR("bad SET PREDICATION\n");
1797 			return -EINVAL;
1798 		}
1799 
1800 		tmp = radeon_get_ib_value(p, idx + 1);
1801 		pred_op = (tmp >> 16) & 0x7;
1802 
1803 		/* for the clear predicate operation */
1804 		if (pred_op == 0)
1805 			return 0;
1806 
1807 		if (pred_op > 2) {
1808 			DRM_ERROR("bad SET PREDICATION operation %d\n", pred_op);
1809 			return -EINVAL;
1810 		}
1811 
1812 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1813 		if (r) {
1814 			DRM_ERROR("bad SET PREDICATION\n");
1815 			return -EINVAL;
1816 		}
1817 
1818 		offset = reloc->gpu_offset +
1819 			 (idx_value & 0xfffffff0) +
1820 			 ((u64)(tmp & 0xff) << 32);
1821 
1822 		ib[idx + 0] = offset;
1823 		ib[idx + 1] = (tmp & 0xffffff00) | (upper_32_bits(offset) & 0xff);
1824 	}
1825 	break;
1826 	case PACKET3_CONTEXT_CONTROL:
1827 		if (pkt->count != 1) {
1828 			DRM_ERROR("bad CONTEXT_CONTROL\n");
1829 			return -EINVAL;
1830 		}
1831 		break;
1832 	case PACKET3_INDEX_TYPE:
1833 	case PACKET3_NUM_INSTANCES:
1834 	case PACKET3_CLEAR_STATE:
1835 		if (pkt->count) {
1836 			DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES/CLEAR_STATE\n");
1837 			return -EINVAL;
1838 		}
1839 		break;
1840 	case CAYMAN_PACKET3_DEALLOC_STATE:
1841 		if (p->rdev->family < CHIP_CAYMAN) {
1842 			DRM_ERROR("bad PACKET3_DEALLOC_STATE\n");
1843 			return -EINVAL;
1844 		}
1845 		if (pkt->count) {
1846 			DRM_ERROR("bad INDEX_TYPE/NUM_INSTANCES/CLEAR_STATE\n");
1847 			return -EINVAL;
1848 		}
1849 		break;
1850 	case PACKET3_INDEX_BASE:
1851 	{
1852 		uint64_t offset;
1853 
1854 		if (pkt->count != 1) {
1855 			DRM_ERROR("bad INDEX_BASE\n");
1856 			return -EINVAL;
1857 		}
1858 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1859 		if (r) {
1860 			DRM_ERROR("bad INDEX_BASE\n");
1861 			return -EINVAL;
1862 		}
1863 
1864 		offset = reloc->gpu_offset +
1865 			 idx_value +
1866 			 ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
1867 
1868 		ib[idx+0] = offset;
1869 		ib[idx+1] = upper_32_bits(offset) & 0xff;
1870 
1871 		r = evergreen_cs_track_check(p);
1872 		if (r) {
1873 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1874 			return r;
1875 		}
1876 		break;
1877 	}
1878 	case PACKET3_INDEX_BUFFER_SIZE:
1879 	{
1880 		if (pkt->count != 0) {
1881 			DRM_ERROR("bad INDEX_BUFFER_SIZE\n");
1882 			return -EINVAL;
1883 		}
1884 		break;
1885 	}
1886 	case PACKET3_DRAW_INDEX:
1887 	{
1888 		uint64_t offset;
1889 		if (pkt->count != 3) {
1890 			DRM_ERROR("bad DRAW_INDEX\n");
1891 			return -EINVAL;
1892 		}
1893 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1894 		if (r) {
1895 			DRM_ERROR("bad DRAW_INDEX\n");
1896 			return -EINVAL;
1897 		}
1898 
1899 		offset = reloc->gpu_offset +
1900 			 idx_value +
1901 			 ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
1902 
1903 		ib[idx+0] = offset;
1904 		ib[idx+1] = upper_32_bits(offset) & 0xff;
1905 
1906 		r = evergreen_cs_track_check(p);
1907 		if (r) {
1908 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1909 			return r;
1910 		}
1911 		break;
1912 	}
1913 	case PACKET3_DRAW_INDEX_2:
1914 	{
1915 		uint64_t offset;
1916 
1917 		if (pkt->count != 4) {
1918 			DRM_ERROR("bad DRAW_INDEX_2\n");
1919 			return -EINVAL;
1920 		}
1921 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1922 		if (r) {
1923 			DRM_ERROR("bad DRAW_INDEX_2\n");
1924 			return -EINVAL;
1925 		}
1926 
1927 		offset = reloc->gpu_offset +
1928 			 radeon_get_ib_value(p, idx+1) +
1929 			 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
1930 
1931 		ib[idx+1] = offset;
1932 		ib[idx+2] = upper_32_bits(offset) & 0xff;
1933 
1934 		r = evergreen_cs_track_check(p);
1935 		if (r) {
1936 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1937 			return r;
1938 		}
1939 		break;
1940 	}
1941 	case PACKET3_DRAW_INDEX_AUTO:
1942 		if (pkt->count != 1) {
1943 			DRM_ERROR("bad DRAW_INDEX_AUTO\n");
1944 			return -EINVAL;
1945 		}
1946 		r = evergreen_cs_track_check(p);
1947 		if (r) {
1948 			dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1949 			return r;
1950 		}
1951 		break;
1952 	case PACKET3_DRAW_INDEX_MULTI_AUTO:
1953 		if (pkt->count != 2) {
1954 			DRM_ERROR("bad DRAW_INDEX_MULTI_AUTO\n");
1955 			return -EINVAL;
1956 		}
1957 		r = evergreen_cs_track_check(p);
1958 		if (r) {
1959 			dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
1960 			return r;
1961 		}
1962 		break;
1963 	case PACKET3_DRAW_INDEX_IMMD:
1964 		if (pkt->count < 2) {
1965 			DRM_ERROR("bad DRAW_INDEX_IMMD\n");
1966 			return -EINVAL;
1967 		}
1968 		r = evergreen_cs_track_check(p);
1969 		if (r) {
1970 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1971 			return r;
1972 		}
1973 		break;
1974 	case PACKET3_DRAW_INDEX_OFFSET:
1975 		if (pkt->count != 2) {
1976 			DRM_ERROR("bad DRAW_INDEX_OFFSET\n");
1977 			return -EINVAL;
1978 		}
1979 		r = evergreen_cs_track_check(p);
1980 		if (r) {
1981 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1982 			return r;
1983 		}
1984 		break;
1985 	case PACKET3_DRAW_INDEX_OFFSET_2:
1986 		if (pkt->count != 3) {
1987 			DRM_ERROR("bad DRAW_INDEX_OFFSET_2\n");
1988 			return -EINVAL;
1989 		}
1990 		r = evergreen_cs_track_check(p);
1991 		if (r) {
1992 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
1993 			return r;
1994 		}
1995 		break;
1996 	case PACKET3_SET_BASE:
1997 	{
1998 		/*
1999 		DW 1 HEADER Header of the packet. Shader_Type in bit 1 of the Header will correspond to the shader type of the Load, see Type-3 Packet.
2000 		   2 BASE_INDEX Bits [3:0] BASE_INDEX - Base Index specifies which base address is specified in the last two DWs.
2001 		     0001: DX11 Draw_Index_Indirect Patch Table Base: Base address for Draw_Index_Indirect data.
2002 		   3 ADDRESS_LO Bits [31:3] - Lower bits of QWORD-Aligned Address. Bits [2:0] - Reserved
2003 		   4 ADDRESS_HI Bits [31:8] - Reserved. Bits [7:0] - Upper bits of Address [47:32]
2004 		*/
2005 		if (pkt->count != 2) {
2006 			DRM_ERROR("bad SET_BASE\n");
2007 			return -EINVAL;
2008 		}
2009 
2010 		/* currently only supporting setting indirect draw buffer base address */
2011 		if (idx_value != 1) {
2012 			DRM_ERROR("bad SET_BASE\n");
2013 			return -EINVAL;
2014 		}
2015 
2016 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2017 		if (r) {
2018 			DRM_ERROR("bad SET_BASE\n");
2019 			return -EINVAL;
2020 		}
2021 
2022 		track->indirect_draw_buffer_size = radeon_bo_size(reloc->robj);
2023 
2024 		ib[idx+1] = reloc->gpu_offset;
2025 		ib[idx+2] = upper_32_bits(reloc->gpu_offset) & 0xff;
2026 
2027 		break;
2028 	}
2029 	case PACKET3_DRAW_INDIRECT:
2030 	case PACKET3_DRAW_INDEX_INDIRECT:
2031 	{
2032 		u64 size = pkt->opcode == PACKET3_DRAW_INDIRECT ? 16 : 20;
2033 
2034 		/*
2035 		DW 1 HEADER
2036 		   2 DATA_OFFSET Bits [31:0] + byte aligned offset where the required data structure starts. Bits 1:0 are zero
2037 		   3 DRAW_INITIATOR Draw Initiator Register. Written to the VGT_DRAW_INITIATOR register for the assigned context
2038 		*/
2039 		if (pkt->count != 1) {
2040 			DRM_ERROR("bad DRAW_INDIRECT\n");
2041 			return -EINVAL;
2042 		}
2043 
2044 		if (idx_value + size > track->indirect_draw_buffer_size) {
2045 			dev_warn(p->dev, "DRAW_INDIRECT buffer too small %u + %llu > %lu\n",
2046 				idx_value, size, track->indirect_draw_buffer_size);
2047 			return -EINVAL;
2048 		}
2049 
2050 		r = evergreen_cs_track_check(p);
2051 		if (r) {
2052 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
2053 			return r;
2054 		}
2055 		break;
2056 	}
2057 	case PACKET3_DISPATCH_DIRECT:
2058 		if (pkt->count != 3) {
2059 			DRM_ERROR("bad DISPATCH_DIRECT\n");
2060 			return -EINVAL;
2061 		}
2062 		r = evergreen_cs_track_check(p);
2063 		if (r) {
2064 			dev_warn(p->dev, "%s:%d invalid cmd stream %d\n", __func__, __LINE__, idx);
2065 			return r;
2066 		}
2067 		break;
2068 	case PACKET3_DISPATCH_INDIRECT:
2069 		if (pkt->count != 1) {
2070 			DRM_ERROR("bad DISPATCH_INDIRECT\n");
2071 			return -EINVAL;
2072 		}
2073 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2074 		if (r) {
2075 			DRM_ERROR("bad DISPATCH_INDIRECT\n");
2076 			return -EINVAL;
2077 		}
2078 		ib[idx+0] = idx_value + (u32)(reloc->gpu_offset & 0xffffffff);
2079 		r = evergreen_cs_track_check(p);
2080 		if (r) {
2081 			dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
2082 			return r;
2083 		}
2084 		break;
2085 	case PACKET3_WAIT_REG_MEM:
2086 		if (pkt->count != 5) {
2087 			DRM_ERROR("bad WAIT_REG_MEM\n");
2088 			return -EINVAL;
2089 		}
2090 		/* bit 4 is reg (0) or mem (1) */
2091 		if (idx_value & 0x10) {
2092 			uint64_t offset;
2093 
2094 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2095 			if (r) {
2096 				DRM_ERROR("bad WAIT_REG_MEM\n");
2097 				return -EINVAL;
2098 			}
2099 
2100 			offset = reloc->gpu_offset +
2101 				 (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
2102 				 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
2103 
2104 			ib[idx+1] = (ib[idx+1] & 0x3) | (offset & 0xfffffffc);
2105 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2106 		} else if (idx_value & 0x100) {
2107 			DRM_ERROR("cannot use PFP on REG wait\n");
2108 			return -EINVAL;
2109 		}
2110 		break;
2111 	case PACKET3_CP_DMA:
2112 	{
2113 		u32 command, size, info;
2114 		u64 offset, tmp;
2115 		if (pkt->count != 4) {
2116 			DRM_ERROR("bad CP DMA\n");
2117 			return -EINVAL;
2118 		}
2119 		command = radeon_get_ib_value(p, idx+4);
2120 		size = command & 0x1fffff;
2121 		info = radeon_get_ib_value(p, idx+1);
2122 		if ((((info & 0x60000000) >> 29) != 0) || /* src = GDS or DATA */
2123 		    (((info & 0x00300000) >> 20) != 0) || /* dst = GDS */
2124 		    ((((info & 0x00300000) >> 20) == 0) &&
2125 		     (command & PACKET3_CP_DMA_CMD_DAS)) || /* dst = register */
2126 		    ((((info & 0x60000000) >> 29) == 0) &&
2127 		     (command & PACKET3_CP_DMA_CMD_SAS))) { /* src = register */
2128 			/* non mem to mem copies requires dw aligned count */
2129 			if (size % 4) {
2130 				DRM_ERROR("CP DMA command requires dw count alignment\n");
2131 				return -EINVAL;
2132 			}
2133 		}
2134 		if (command & PACKET3_CP_DMA_CMD_SAS) {
2135 			/* src address space is register */
2136 			/* GDS is ok */
2137 			if (((info & 0x60000000) >> 29) != 1) {
2138 				DRM_ERROR("CP DMA SAS not supported\n");
2139 				return -EINVAL;
2140 			}
2141 		} else {
2142 			if (command & PACKET3_CP_DMA_CMD_SAIC) {
2143 				DRM_ERROR("CP DMA SAIC only supported for registers\n");
2144 				return -EINVAL;
2145 			}
2146 			/* src address space is memory */
2147 			if (((info & 0x60000000) >> 29) == 0) {
2148 				r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2149 				if (r) {
2150 					DRM_ERROR("bad CP DMA SRC\n");
2151 					return -EINVAL;
2152 				}
2153 
2154 				tmp = radeon_get_ib_value(p, idx) +
2155 					((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
2156 
2157 				offset = reloc->gpu_offset + tmp;
2158 
2159 				if ((tmp + size) > radeon_bo_size(reloc->robj)) {
2160 					dev_warn(p->dev, "CP DMA src buffer too small (%llu %lu)\n",
2161 						 tmp + size, radeon_bo_size(reloc->robj));
2162 					return -EINVAL;
2163 				}
2164 
2165 				ib[idx] = offset;
2166 				ib[idx+1] = (ib[idx+1] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
2167 			} else if (((info & 0x60000000) >> 29) != 2) {
2168 				DRM_ERROR("bad CP DMA SRC_SEL\n");
2169 				return -EINVAL;
2170 			}
2171 		}
2172 		if (command & PACKET3_CP_DMA_CMD_DAS) {
2173 			/* dst address space is register */
2174 			/* GDS is ok */
2175 			if (((info & 0x00300000) >> 20) != 1) {
2176 				DRM_ERROR("CP DMA DAS not supported\n");
2177 				return -EINVAL;
2178 			}
2179 		} else {
2180 			/* dst address space is memory */
2181 			if (command & PACKET3_CP_DMA_CMD_DAIC) {
2182 				DRM_ERROR("CP DMA DAIC only supported for registers\n");
2183 				return -EINVAL;
2184 			}
2185 			if (((info & 0x00300000) >> 20) == 0) {
2186 				r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2187 				if (r) {
2188 					DRM_ERROR("bad CP DMA DST\n");
2189 					return -EINVAL;
2190 				}
2191 
2192 				tmp = radeon_get_ib_value(p, idx+2) +
2193 					((u64)(radeon_get_ib_value(p, idx+3) & 0xff) << 32);
2194 
2195 				offset = reloc->gpu_offset + tmp;
2196 
2197 				if ((tmp + size) > radeon_bo_size(reloc->robj)) {
2198 					dev_warn(p->dev, "CP DMA dst buffer too small (%llu %lu)\n",
2199 						 tmp + size, radeon_bo_size(reloc->robj));
2200 					return -EINVAL;
2201 				}
2202 
2203 				ib[idx+2] = offset;
2204 				ib[idx+3] = upper_32_bits(offset) & 0xff;
2205 			} else {
2206 				DRM_ERROR("bad CP DMA DST_SEL\n");
2207 				return -EINVAL;
2208 			}
2209 		}
2210 		break;
2211 	}
2212 	case PACKET3_PFP_SYNC_ME:
2213 		if (pkt->count) {
2214 			DRM_ERROR("bad PFP_SYNC_ME\n");
2215 			return -EINVAL;
2216 		}
2217 		break;
2218 	case PACKET3_SURFACE_SYNC:
2219 		if (pkt->count != 3) {
2220 			DRM_ERROR("bad SURFACE_SYNC\n");
2221 			return -EINVAL;
2222 		}
2223 		/* 0xffffffff/0x0 is flush all cache flag */
2224 		if (radeon_get_ib_value(p, idx + 1) != 0xffffffff ||
2225 		    radeon_get_ib_value(p, idx + 2) != 0) {
2226 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2227 			if (r) {
2228 				DRM_ERROR("bad SURFACE_SYNC\n");
2229 				return -EINVAL;
2230 			}
2231 			ib[idx+2] += (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
2232 		}
2233 		break;
2234 	case PACKET3_EVENT_WRITE:
2235 		if (pkt->count != 2 && pkt->count != 0) {
2236 			DRM_ERROR("bad EVENT_WRITE\n");
2237 			return -EINVAL;
2238 		}
2239 		if (pkt->count) {
2240 			uint64_t offset;
2241 
2242 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2243 			if (r) {
2244 				DRM_ERROR("bad EVENT_WRITE\n");
2245 				return -EINVAL;
2246 			}
2247 			offset = reloc->gpu_offset +
2248 				 (radeon_get_ib_value(p, idx+1) & 0xfffffff8) +
2249 				 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
2250 
2251 			ib[idx+1] = offset & 0xfffffff8;
2252 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2253 		}
2254 		break;
2255 	case PACKET3_EVENT_WRITE_EOP:
2256 	{
2257 		uint64_t offset;
2258 
2259 		if (pkt->count != 4) {
2260 			DRM_ERROR("bad EVENT_WRITE_EOP\n");
2261 			return -EINVAL;
2262 		}
2263 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2264 		if (r) {
2265 			DRM_ERROR("bad EVENT_WRITE_EOP\n");
2266 			return -EINVAL;
2267 		}
2268 
2269 		offset = reloc->gpu_offset +
2270 			 (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
2271 			 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
2272 
2273 		ib[idx+1] = offset & 0xfffffffc;
2274 		ib[idx+2] = (ib[idx+2] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
2275 		break;
2276 	}
2277 	case PACKET3_EVENT_WRITE_EOS:
2278 	{
2279 		uint64_t offset;
2280 
2281 		if (pkt->count != 3) {
2282 			DRM_ERROR("bad EVENT_WRITE_EOS\n");
2283 			return -EINVAL;
2284 		}
2285 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2286 		if (r) {
2287 			DRM_ERROR("bad EVENT_WRITE_EOS\n");
2288 			return -EINVAL;
2289 		}
2290 
2291 		offset = reloc->gpu_offset +
2292 			 (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
2293 			 ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
2294 
2295 		ib[idx+1] = offset & 0xfffffffc;
2296 		ib[idx+2] = (ib[idx+2] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
2297 		break;
2298 	}
2299 	case PACKET3_SET_CONFIG_REG:
2300 		start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START;
2301 		end_reg = 4 * pkt->count + start_reg - 4;
2302 		if ((start_reg < PACKET3_SET_CONFIG_REG_START) ||
2303 		    (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
2304 		    (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
2305 			DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
2306 			return -EINVAL;
2307 		}
2308 		for (reg = start_reg, idx++; reg <= end_reg; reg += 4, idx++) {
2309 			if (evergreen_is_safe_reg(p, reg))
2310 				continue;
2311 			r = evergreen_cs_handle_reg(p, reg, idx);
2312 			if (r)
2313 				return r;
2314 		}
2315 		break;
2316 	case PACKET3_SET_CONTEXT_REG:
2317 		start_reg = (idx_value << 2) + PACKET3_SET_CONTEXT_REG_START;
2318 		end_reg = 4 * pkt->count + start_reg - 4;
2319 		if ((start_reg < PACKET3_SET_CONTEXT_REG_START) ||
2320 		    (start_reg >= PACKET3_SET_CONTEXT_REG_END) ||
2321 		    (end_reg >= PACKET3_SET_CONTEXT_REG_END)) {
2322 			DRM_ERROR("bad PACKET3_SET_CONTEXT_REG\n");
2323 			return -EINVAL;
2324 		}
2325 		for (reg = start_reg, idx++; reg <= end_reg; reg += 4, idx++) {
2326 			if (evergreen_is_safe_reg(p, reg))
2327 				continue;
2328 			r = evergreen_cs_handle_reg(p, reg, idx);
2329 			if (r)
2330 				return r;
2331 		}
2332 		break;
2333 	case PACKET3_SET_RESOURCE:
2334 		if (pkt->count % 8) {
2335 			DRM_ERROR("bad SET_RESOURCE\n");
2336 			return -EINVAL;
2337 		}
2338 		start_reg = (idx_value << 2) + PACKET3_SET_RESOURCE_START;
2339 		end_reg = 4 * pkt->count + start_reg - 4;
2340 		if ((start_reg < PACKET3_SET_RESOURCE_START) ||
2341 		    (start_reg >= PACKET3_SET_RESOURCE_END) ||
2342 		    (end_reg >= PACKET3_SET_RESOURCE_END)) {
2343 			DRM_ERROR("bad SET_RESOURCE\n");
2344 			return -EINVAL;
2345 		}
2346 		for (i = 0; i < (pkt->count / 8); i++) {
2347 			struct radeon_bo *texture, *mipmap;
2348 			u32 toffset, moffset;
2349 			u32 size, offset, mip_address, tex_dim;
2350 
2351 			switch (G__SQ_CONSTANT_TYPE(radeon_get_ib_value(p, idx+1+(i*8)+7))) {
2352 			case SQ_TEX_VTX_VALID_TEXTURE:
2353 				/* tex base */
2354 				r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2355 				if (r) {
2356 					DRM_ERROR("bad SET_RESOURCE (tex)\n");
2357 					return -EINVAL;
2358 				}
2359 				if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
2360 					ib[idx+1+(i*8)+1] |=
2361 						TEX_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->tiling_flags));
2362 					if (reloc->tiling_flags & RADEON_TILING_MACRO) {
2363 						unsigned bankw, bankh, mtaspect, tile_split;
2364 
2365 						evergreen_tiling_fields(reloc->tiling_flags,
2366 									&bankw, &bankh, &mtaspect,
2367 									&tile_split);
2368 						ib[idx+1+(i*8)+6] |= TEX_TILE_SPLIT(tile_split);
2369 						ib[idx+1+(i*8)+7] |=
2370 							TEX_BANK_WIDTH(bankw) |
2371 							TEX_BANK_HEIGHT(bankh) |
2372 							MACRO_TILE_ASPECT(mtaspect) |
2373 							TEX_NUM_BANKS(evergreen_cs_get_num_banks(track->nbanks));
2374 					}
2375 				}
2376 				texture = reloc->robj;
2377 				toffset = (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
2378 
2379 				/* tex mip base */
2380 				tex_dim = ib[idx+1+(i*8)+0] & 0x7;
2381 				mip_address = ib[idx+1+(i*8)+3];
2382 
2383 				if ((tex_dim == SQ_TEX_DIM_2D_MSAA || tex_dim == SQ_TEX_DIM_2D_ARRAY_MSAA) &&
2384 				    !mip_address &&
2385 				    !radeon_cs_packet_next_is_pkt3_nop(p)) {
2386 					/* MIP_ADDRESS should point to FMASK for an MSAA texture.
2387 					 * It should be 0 if FMASK is disabled. */
2388 					moffset = 0;
2389 					mipmap = NULL;
2390 				} else {
2391 					r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2392 					if (r) {
2393 						DRM_ERROR("bad SET_RESOURCE (tex)\n");
2394 						return -EINVAL;
2395 					}
2396 					moffset = (u32)((reloc->gpu_offset >> 8) & 0xffffffff);
2397 					mipmap = reloc->robj;
2398 				}
2399 
2400 				r = evergreen_cs_track_validate_texture(p, texture, mipmap, idx+1+(i*8));
2401 				if (r)
2402 					return r;
2403 				ib[idx+1+(i*8)+2] += toffset;
2404 				ib[idx+1+(i*8)+3] += moffset;
2405 				break;
2406 			case SQ_TEX_VTX_VALID_BUFFER:
2407 			{
2408 				uint64_t offset64;
2409 				/* vtx base */
2410 				r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2411 				if (r) {
2412 					DRM_ERROR("bad SET_RESOURCE (vtx)\n");
2413 					return -EINVAL;
2414 				}
2415 				offset = radeon_get_ib_value(p, idx+1+(i*8)+0);
2416 				size = radeon_get_ib_value(p, idx+1+(i*8)+1);
2417 				if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
2418 					/* force size to size of the buffer */
2419 					dev_warn_ratelimited(p->dev, "vbo resource seems too big for the bo\n");
2420 					ib[idx+1+(i*8)+1] = radeon_bo_size(reloc->robj) - offset;
2421 				}
2422 
2423 				offset64 = reloc->gpu_offset + offset;
2424 				ib[idx+1+(i*8)+0] = offset64;
2425 				ib[idx+1+(i*8)+2] = (ib[idx+1+(i*8)+2] & 0xffffff00) |
2426 						    (upper_32_bits(offset64) & 0xff);
2427 				break;
2428 			}
2429 			case SQ_TEX_VTX_INVALID_TEXTURE:
2430 			case SQ_TEX_VTX_INVALID_BUFFER:
2431 			default:
2432 				DRM_ERROR("bad SET_RESOURCE\n");
2433 				return -EINVAL;
2434 			}
2435 		}
2436 		break;
2437 	case PACKET3_SET_ALU_CONST:
2438 		/* XXX fix me ALU const buffers only */
2439 		break;
2440 	case PACKET3_SET_BOOL_CONST:
2441 		start_reg = (idx_value << 2) + PACKET3_SET_BOOL_CONST_START;
2442 		end_reg = 4 * pkt->count + start_reg - 4;
2443 		if ((start_reg < PACKET3_SET_BOOL_CONST_START) ||
2444 		    (start_reg >= PACKET3_SET_BOOL_CONST_END) ||
2445 		    (end_reg >= PACKET3_SET_BOOL_CONST_END)) {
2446 			DRM_ERROR("bad SET_BOOL_CONST\n");
2447 			return -EINVAL;
2448 		}
2449 		break;
2450 	case PACKET3_SET_LOOP_CONST:
2451 		start_reg = (idx_value << 2) + PACKET3_SET_LOOP_CONST_START;
2452 		end_reg = 4 * pkt->count + start_reg - 4;
2453 		if ((start_reg < PACKET3_SET_LOOP_CONST_START) ||
2454 		    (start_reg >= PACKET3_SET_LOOP_CONST_END) ||
2455 		    (end_reg >= PACKET3_SET_LOOP_CONST_END)) {
2456 			DRM_ERROR("bad SET_LOOP_CONST\n");
2457 			return -EINVAL;
2458 		}
2459 		break;
2460 	case PACKET3_SET_CTL_CONST:
2461 		start_reg = (idx_value << 2) + PACKET3_SET_CTL_CONST_START;
2462 		end_reg = 4 * pkt->count + start_reg - 4;
2463 		if ((start_reg < PACKET3_SET_CTL_CONST_START) ||
2464 		    (start_reg >= PACKET3_SET_CTL_CONST_END) ||
2465 		    (end_reg >= PACKET3_SET_CTL_CONST_END)) {
2466 			DRM_ERROR("bad SET_CTL_CONST\n");
2467 			return -EINVAL;
2468 		}
2469 		break;
2470 	case PACKET3_SET_SAMPLER:
2471 		if (pkt->count % 3) {
2472 			DRM_ERROR("bad SET_SAMPLER\n");
2473 			return -EINVAL;
2474 		}
2475 		start_reg = (idx_value << 2) + PACKET3_SET_SAMPLER_START;
2476 		end_reg = 4 * pkt->count + start_reg - 4;
2477 		if ((start_reg < PACKET3_SET_SAMPLER_START) ||
2478 		    (start_reg >= PACKET3_SET_SAMPLER_END) ||
2479 		    (end_reg >= PACKET3_SET_SAMPLER_END)) {
2480 			DRM_ERROR("bad SET_SAMPLER\n");
2481 			return -EINVAL;
2482 		}
2483 		break;
2484 	case PACKET3_STRMOUT_BUFFER_UPDATE:
2485 		if (pkt->count != 4) {
2486 			DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (invalid count)\n");
2487 			return -EINVAL;
2488 		}
2489 		/* Updating memory at DST_ADDRESS. */
2490 		if (idx_value & 0x1) {
2491 			u64 offset;
2492 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2493 			if (r) {
2494 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing dst reloc)\n");
2495 				return -EINVAL;
2496 			}
2497 			offset = radeon_get_ib_value(p, idx+1);
2498 			offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2499 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2500 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE dst bo too small: 0x%llx, 0x%lx\n",
2501 					  offset + 4, radeon_bo_size(reloc->robj));
2502 				return -EINVAL;
2503 			}
2504 			offset += reloc->gpu_offset;
2505 			ib[idx+1] = offset;
2506 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2507 		}
2508 		/* Reading data from SRC_ADDRESS. */
2509 		if (((idx_value >> 1) & 0x3) == 2) {
2510 			u64 offset;
2511 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2512 			if (r) {
2513 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE (missing src reloc)\n");
2514 				return -EINVAL;
2515 			}
2516 			offset = radeon_get_ib_value(p, idx+3);
2517 			offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2518 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2519 				DRM_ERROR("bad STRMOUT_BUFFER_UPDATE src bo too small: 0x%llx, 0x%lx\n",
2520 					  offset + 4, radeon_bo_size(reloc->robj));
2521 				return -EINVAL;
2522 			}
2523 			offset += reloc->gpu_offset;
2524 			ib[idx+3] = offset;
2525 			ib[idx+4] = upper_32_bits(offset) & 0xff;
2526 		}
2527 		break;
2528 	case PACKET3_MEM_WRITE:
2529 	{
2530 		u64 offset;
2531 
2532 		if (pkt->count != 3) {
2533 			DRM_ERROR("bad MEM_WRITE (invalid count)\n");
2534 			return -EINVAL;
2535 		}
2536 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2537 		if (r) {
2538 			DRM_ERROR("bad MEM_WRITE (missing reloc)\n");
2539 			return -EINVAL;
2540 		}
2541 		offset = radeon_get_ib_value(p, idx+0);
2542 		offset += ((u64)(radeon_get_ib_value(p, idx+1) & 0xff)) << 32UL;
2543 		if (offset & 0x7) {
2544 			DRM_ERROR("bad MEM_WRITE (address not qwords aligned)\n");
2545 			return -EINVAL;
2546 		}
2547 		if ((offset + 8) > radeon_bo_size(reloc->robj)) {
2548 			DRM_ERROR("bad MEM_WRITE bo too small: 0x%llx, 0x%lx\n",
2549 				  offset + 8, radeon_bo_size(reloc->robj));
2550 			return -EINVAL;
2551 		}
2552 		offset += reloc->gpu_offset;
2553 		ib[idx+0] = offset;
2554 		ib[idx+1] = upper_32_bits(offset) & 0xff;
2555 		break;
2556 	}
2557 	case PACKET3_COPY_DW:
2558 		if (pkt->count != 4) {
2559 			DRM_ERROR("bad COPY_DW (invalid count)\n");
2560 			return -EINVAL;
2561 		}
2562 		if (idx_value & 0x1) {
2563 			u64 offset;
2564 			/* SRC is memory. */
2565 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2566 			if (r) {
2567 				DRM_ERROR("bad COPY_DW (missing src reloc)\n");
2568 				return -EINVAL;
2569 			}
2570 			offset = radeon_get_ib_value(p, idx+1);
2571 			offset += ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2572 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2573 				DRM_ERROR("bad COPY_DW src bo too small: 0x%llx, 0x%lx\n",
2574 					  offset + 4, radeon_bo_size(reloc->robj));
2575 				return -EINVAL;
2576 			}
2577 			offset += reloc->gpu_offset;
2578 			ib[idx+1] = offset;
2579 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2580 		} else {
2581 			/* SRC is a reg. */
2582 			reg = radeon_get_ib_value(p, idx+1) << 2;
2583 			if (!evergreen_is_safe_reg(p, reg)) {
2584 				dev_warn(p->dev, "forbidden register 0x%08x at %d\n",
2585 					 reg, idx + 1);
2586 				return -EINVAL;
2587 			}
2588 		}
2589 		if (idx_value & 0x2) {
2590 			u64 offset;
2591 			/* DST is memory. */
2592 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2593 			if (r) {
2594 				DRM_ERROR("bad COPY_DW (missing dst reloc)\n");
2595 				return -EINVAL;
2596 			}
2597 			offset = radeon_get_ib_value(p, idx+3);
2598 			offset += ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2599 			if ((offset + 4) > radeon_bo_size(reloc->robj)) {
2600 				DRM_ERROR("bad COPY_DW dst bo too small: 0x%llx, 0x%lx\n",
2601 					  offset + 4, radeon_bo_size(reloc->robj));
2602 				return -EINVAL;
2603 			}
2604 			offset += reloc->gpu_offset;
2605 			ib[idx+3] = offset;
2606 			ib[idx+4] = upper_32_bits(offset) & 0xff;
2607 		} else {
2608 			/* DST is a reg. */
2609 			reg = radeon_get_ib_value(p, idx+3) << 2;
2610 			if (!evergreen_is_safe_reg(p, reg)) {
2611 				dev_warn(p->dev, "forbidden register 0x%08x at %d\n",
2612 					 reg, idx + 3);
2613 				return -EINVAL;
2614 			}
2615 		}
2616 		break;
2617 	case PACKET3_SET_APPEND_CNT:
2618 	{
2619 		uint32_t areg;
2620 		uint32_t allowed_reg_base;
2621 		uint32_t source_sel;
2622 		if (pkt->count != 2) {
2623 			DRM_ERROR("bad SET_APPEND_CNT (invalid count)\n");
2624 			return -EINVAL;
2625 		}
2626 
2627 		allowed_reg_base = GDS_APPEND_COUNT_0;
2628 		allowed_reg_base -= PACKET3_SET_CONTEXT_REG_START;
2629 		allowed_reg_base >>= 2;
2630 
2631 		areg = idx_value >> 16;
2632 		if (areg < allowed_reg_base || areg > (allowed_reg_base + 11)) {
2633 			dev_warn(p->dev, "forbidden register for append cnt 0x%08x at %d\n",
2634 				 areg, idx);
2635 			return -EINVAL;
2636 		}
2637 
2638 		source_sel = G_PACKET3_SET_APPEND_CNT_SRC_SELECT(idx_value);
2639 		if (source_sel == PACKET3_SAC_SRC_SEL_MEM) {
2640 			uint64_t offset;
2641 			uint32_t swap;
2642 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
2643 			if (r) {
2644 				DRM_ERROR("bad SET_APPEND_CNT (missing reloc)\n");
2645 				return -EINVAL;
2646 			}
2647 			offset = radeon_get_ib_value(p, idx + 1);
2648 			swap = offset & 0x3;
2649 			offset &= ~0x3;
2650 
2651 			offset += ((u64)(radeon_get_ib_value(p, idx + 2) & 0xff)) << 32;
2652 
2653 			offset += reloc->gpu_offset;
2654 			ib[idx+1] = (offset & 0xfffffffc) | swap;
2655 			ib[idx+2] = upper_32_bits(offset) & 0xff;
2656 		} else {
2657 			DRM_ERROR("bad SET_APPEND_CNT (unsupported operation)\n");
2658 			return -EINVAL;
2659 		}
2660 		break;
2661 	}
2662 	case PACKET3_NOP:
2663 		break;
2664 	default:
2665 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2666 		return -EINVAL;
2667 	}
2668 	return 0;
2669 }
2670 
2671 int evergreen_cs_parse(struct radeon_cs_parser *p)
2672 {
2673 	struct radeon_cs_packet pkt;
2674 	struct evergreen_cs_track *track;
2675 	u32 tmp;
2676 	int r;
2677 
2678 	if (p->track == NULL) {
2679 		/* initialize tracker, we are in kms */
2680 		track = kzalloc(sizeof(*track), GFP_KERNEL);
2681 		if (track == NULL)
2682 			return -ENOMEM;
2683 		evergreen_cs_track_init(track);
2684 		if (p->rdev->family >= CHIP_CAYMAN) {
2685 			tmp = p->rdev->config.cayman.tile_config;
2686 			track->reg_safe_bm = cayman_reg_safe_bm;
2687 		} else {
2688 			tmp = p->rdev->config.evergreen.tile_config;
2689 			track->reg_safe_bm = evergreen_reg_safe_bm;
2690 		}
2691 		BUILD_BUG_ON(ARRAY_SIZE(cayman_reg_safe_bm) != REG_SAFE_BM_SIZE);
2692 		BUILD_BUG_ON(ARRAY_SIZE(evergreen_reg_safe_bm) != REG_SAFE_BM_SIZE);
2693 		switch (tmp & 0xf) {
2694 		case 0:
2695 			track->npipes = 1;
2696 			break;
2697 		case 1:
2698 		default:
2699 			track->npipes = 2;
2700 			break;
2701 		case 2:
2702 			track->npipes = 4;
2703 			break;
2704 		case 3:
2705 			track->npipes = 8;
2706 			break;
2707 		}
2708 
2709 		switch ((tmp & 0xf0) >> 4) {
2710 		case 0:
2711 			track->nbanks = 4;
2712 			break;
2713 		case 1:
2714 		default:
2715 			track->nbanks = 8;
2716 			break;
2717 		case 2:
2718 			track->nbanks = 16;
2719 			break;
2720 		}
2721 
2722 		switch ((tmp & 0xf00) >> 8) {
2723 		case 0:
2724 			track->group_size = 256;
2725 			break;
2726 		case 1:
2727 		default:
2728 			track->group_size = 512;
2729 			break;
2730 		}
2731 
2732 		switch ((tmp & 0xf000) >> 12) {
2733 		case 0:
2734 			track->row_size = 1;
2735 			break;
2736 		case 1:
2737 		default:
2738 			track->row_size = 2;
2739 			break;
2740 		case 2:
2741 			track->row_size = 4;
2742 			break;
2743 		}
2744 
2745 		p->track = track;
2746 	}
2747 	do {
2748 		r = radeon_cs_packet_parse(p, &pkt, p->idx);
2749 		if (r) {
2750 			kfree(p->track);
2751 			p->track = NULL;
2752 			return r;
2753 		}
2754 		p->idx += pkt.count + 2;
2755 		switch (pkt.type) {
2756 		case RADEON_PACKET_TYPE0:
2757 			r = evergreen_cs_parse_packet0(p, &pkt);
2758 			break;
2759 		case RADEON_PACKET_TYPE2:
2760 			break;
2761 		case RADEON_PACKET_TYPE3:
2762 			r = evergreen_packet3_check(p, &pkt);
2763 			break;
2764 		default:
2765 			DRM_ERROR("Unknown packet type %d !\n", pkt.type);
2766 			kfree(p->track);
2767 			p->track = NULL;
2768 			return -EINVAL;
2769 		}
2770 		if (r) {
2771 			kfree(p->track);
2772 			p->track = NULL;
2773 			return r;
2774 		}
2775 	} while (p->idx < p->chunk_ib->length_dw);
2776 #if 0
2777 	for (r = 0; r < p->ib.length_dw; r++) {
2778 		pr_info("%05d  0x%08X\n", r, p->ib.ptr[r]);
2779 		mdelay(1);
2780 	}
2781 #endif
2782 	kfree(p->track);
2783 	p->track = NULL;
2784 	return 0;
2785 }
2786 
2787 /**
2788  * evergreen_dma_cs_parse() - parse the DMA IB
2789  * @p:		parser structure holding parsing context.
2790  *
2791  * Parses the DMA IB from the CS ioctl and updates
2792  * the GPU addresses based on the reloc information and
2793  * checks for errors. (Evergreen-Cayman)
2794  * Returns 0 for success and an error on failure.
2795  **/
2796 int evergreen_dma_cs_parse(struct radeon_cs_parser *p)
2797 {
2798 	struct radeon_cs_chunk *ib_chunk = p->chunk_ib;
2799 	struct radeon_bo_list *src_reloc, *dst_reloc, *dst2_reloc;
2800 	u32 header, cmd, count, sub_cmd;
2801 	uint32_t *ib = p->ib.ptr;
2802 	u32 idx;
2803 	u64 src_offset, dst_offset, dst2_offset;
2804 	int r;
2805 
2806 	do {
2807 		if (p->idx >= ib_chunk->length_dw) {
2808 			DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
2809 				  p->idx, ib_chunk->length_dw);
2810 			return -EINVAL;
2811 		}
2812 		idx = p->idx;
2813 		header = radeon_get_ib_value(p, idx);
2814 		cmd = GET_DMA_CMD(header);
2815 		count = GET_DMA_COUNT(header);
2816 		sub_cmd = GET_DMA_SUB_CMD(header);
2817 
2818 		switch (cmd) {
2819 		case DMA_PACKET_WRITE:
2820 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
2821 			if (r) {
2822 				DRM_ERROR("bad DMA_PACKET_WRITE\n");
2823 				return -EINVAL;
2824 			}
2825 			switch (sub_cmd) {
2826 			/* tiled */
2827 			case 8:
2828 				dst_offset = radeon_get_ib_value(p, idx+1);
2829 				dst_offset <<= 8;
2830 
2831 				ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
2832 				p->idx += count + 7;
2833 				break;
2834 			/* linear */
2835 			case 0:
2836 				dst_offset = radeon_get_ib_value(p, idx+1);
2837 				dst_offset |= ((u64)(radeon_get_ib_value(p, idx+2) & 0xff)) << 32;
2838 
2839 				ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2840 				ib[idx+2] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2841 				p->idx += count + 3;
2842 				break;
2843 			default:
2844 				DRM_ERROR("bad DMA_PACKET_WRITE [%6d] 0x%08x sub cmd is not 0 or 8\n", idx, header);
2845 				return -EINVAL;
2846 			}
2847 			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2848 				dev_warn(p->dev, "DMA write buffer too small (%llu %lu)\n",
2849 					 dst_offset, radeon_bo_size(dst_reloc->robj));
2850 				return -EINVAL;
2851 			}
2852 			break;
2853 		case DMA_PACKET_COPY:
2854 			r = r600_dma_cs_next_reloc(p, &src_reloc);
2855 			if (r) {
2856 				DRM_ERROR("bad DMA_PACKET_COPY\n");
2857 				return -EINVAL;
2858 			}
2859 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
2860 			if (r) {
2861 				DRM_ERROR("bad DMA_PACKET_COPY\n");
2862 				return -EINVAL;
2863 			}
2864 			switch (sub_cmd) {
2865 			/* Copy L2L, DW aligned */
2866 			case 0x00:
2867 				/* L2L, dw */
2868 				src_offset = radeon_get_ib_value(p, idx+2);
2869 				src_offset |= ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2870 				dst_offset = radeon_get_ib_value(p, idx+1);
2871 				dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
2872 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
2873 					dev_warn(p->dev, "DMA L2L, dw src buffer too small (%llu %lu)\n",
2874 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
2875 					return -EINVAL;
2876 				}
2877 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2878 					dev_warn(p->dev, "DMA L2L, dw dst buffer too small (%llu %lu)\n",
2879 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2880 					return -EINVAL;
2881 				}
2882 				ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2883 				ib[idx+2] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2884 				ib[idx+3] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2885 				ib[idx+4] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2886 				p->idx += 5;
2887 				break;
2888 			/* Copy L2T/T2L */
2889 			case 0x08:
2890 				/* detile bit */
2891 				if (radeon_get_ib_value(p, idx + 2) & (1 << 31)) {
2892 					/* tiled src, linear dst */
2893 					src_offset = radeon_get_ib_value(p, idx+1);
2894 					src_offset <<= 8;
2895 					ib[idx+1] += (u32)(src_reloc->gpu_offset >> 8);
2896 
2897 					dst_offset = radeon_get_ib_value(p, idx + 7);
2898 					dst_offset |= ((u64)(radeon_get_ib_value(p, idx+8) & 0xff)) << 32;
2899 					ib[idx+7] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2900 					ib[idx+8] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2901 				} else {
2902 					/* linear src, tiled dst */
2903 					src_offset = radeon_get_ib_value(p, idx+7);
2904 					src_offset |= ((u64)(radeon_get_ib_value(p, idx+8) & 0xff)) << 32;
2905 					ib[idx+7] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2906 					ib[idx+8] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2907 
2908 					dst_offset = radeon_get_ib_value(p, idx+1);
2909 					dst_offset <<= 8;
2910 					ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
2911 				}
2912 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
2913 					dev_warn(p->dev, "DMA L2T, src buffer too small (%llu %lu)\n",
2914 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
2915 					return -EINVAL;
2916 				}
2917 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2918 					dev_warn(p->dev, "DMA L2T, dst buffer too small (%llu %lu)\n",
2919 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2920 					return -EINVAL;
2921 				}
2922 				p->idx += 9;
2923 				break;
2924 			/* Copy L2L, byte aligned */
2925 			case 0x40:
2926 				/* L2L, byte */
2927 				src_offset = radeon_get_ib_value(p, idx+2);
2928 				src_offset |= ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2929 				dst_offset = radeon_get_ib_value(p, idx+1);
2930 				dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0xff)) << 32;
2931 				if ((src_offset + count) > radeon_bo_size(src_reloc->robj)) {
2932 					dev_warn(p->dev, "DMA L2L, byte src buffer too small (%llu %lu)\n",
2933 							src_offset + count, radeon_bo_size(src_reloc->robj));
2934 					return -EINVAL;
2935 				}
2936 				if ((dst_offset + count) > radeon_bo_size(dst_reloc->robj)) {
2937 					dev_warn(p->dev, "DMA L2L, byte dst buffer too small (%llu %lu)\n",
2938 							dst_offset + count, radeon_bo_size(dst_reloc->robj));
2939 					return -EINVAL;
2940 				}
2941 				ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xffffffff);
2942 				ib[idx+2] += (u32)(src_reloc->gpu_offset & 0xffffffff);
2943 				ib[idx+3] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2944 				ib[idx+4] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2945 				p->idx += 5;
2946 				break;
2947 			/* Copy L2L, partial */
2948 			case 0x41:
2949 				/* L2L, partial */
2950 				if (p->family < CHIP_CAYMAN) {
2951 					DRM_ERROR("L2L Partial is cayman only !\n");
2952 					return -EINVAL;
2953 				}
2954 				ib[idx+1] += (u32)(src_reloc->gpu_offset & 0xffffffff);
2955 				ib[idx+2] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2956 				ib[idx+4] += (u32)(dst_reloc->gpu_offset & 0xffffffff);
2957 				ib[idx+5] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2958 
2959 				p->idx += 9;
2960 				break;
2961 			/* Copy L2L, DW aligned, broadcast */
2962 			case 0x44:
2963 				/* L2L, dw, broadcast */
2964 				r = r600_dma_cs_next_reloc(p, &dst2_reloc);
2965 				if (r) {
2966 					DRM_ERROR("bad L2L, dw, broadcast DMA_PACKET_COPY\n");
2967 					return -EINVAL;
2968 				}
2969 				dst_offset = radeon_get_ib_value(p, idx+1);
2970 				dst_offset |= ((u64)(radeon_get_ib_value(p, idx+4) & 0xff)) << 32;
2971 				dst2_offset = radeon_get_ib_value(p, idx+2);
2972 				dst2_offset |= ((u64)(radeon_get_ib_value(p, idx+5) & 0xff)) << 32;
2973 				src_offset = radeon_get_ib_value(p, idx+3);
2974 				src_offset |= ((u64)(radeon_get_ib_value(p, idx+6) & 0xff)) << 32;
2975 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
2976 					dev_warn(p->dev, "DMA L2L, dw, broadcast src buffer too small (%llu %lu)\n",
2977 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
2978 					return -EINVAL;
2979 				}
2980 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
2981 					dev_warn(p->dev, "DMA L2L, dw, broadcast dst buffer too small (%llu %lu)\n",
2982 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
2983 					return -EINVAL;
2984 				}
2985 				if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
2986 					dev_warn(p->dev, "DMA L2L, dw, broadcast dst2 buffer too small (%llu %lu)\n",
2987 							dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
2988 					return -EINVAL;
2989 				}
2990 				ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
2991 				ib[idx+2] += (u32)(dst2_reloc->gpu_offset & 0xfffffffc);
2992 				ib[idx+3] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
2993 				ib[idx+4] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
2994 				ib[idx+5] += upper_32_bits(dst2_reloc->gpu_offset) & 0xff;
2995 				ib[idx+6] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
2996 				p->idx += 7;
2997 				break;
2998 			/* Copy L2T Frame to Field */
2999 			case 0x48:
3000 				if (radeon_get_ib_value(p, idx + 2) & (1 << 31)) {
3001 					DRM_ERROR("bad L2T, frame to fields DMA_PACKET_COPY\n");
3002 					return -EINVAL;
3003 				}
3004 				r = r600_dma_cs_next_reloc(p, &dst2_reloc);
3005 				if (r) {
3006 					DRM_ERROR("bad L2T, frame to fields DMA_PACKET_COPY\n");
3007 					return -EINVAL;
3008 				}
3009 				dst_offset = radeon_get_ib_value(p, idx+1);
3010 				dst_offset <<= 8;
3011 				dst2_offset = radeon_get_ib_value(p, idx+2);
3012 				dst2_offset <<= 8;
3013 				src_offset = radeon_get_ib_value(p, idx+8);
3014 				src_offset |= ((u64)(radeon_get_ib_value(p, idx+9) & 0xff)) << 32;
3015 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3016 					dev_warn(p->dev, "DMA L2T, frame to fields src buffer too small (%llu %lu)\n",
3017 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3018 					return -EINVAL;
3019 				}
3020 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3021 					dev_warn(p->dev, "DMA L2T, frame to fields buffer too small (%llu %lu)\n",
3022 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3023 					return -EINVAL;
3024 				}
3025 				if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
3026 					dev_warn(p->dev, "DMA L2T, frame to fields buffer too small (%llu %lu)\n",
3027 							dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
3028 					return -EINVAL;
3029 				}
3030 				ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
3031 				ib[idx+2] += (u32)(dst2_reloc->gpu_offset >> 8);
3032 				ib[idx+8] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
3033 				ib[idx+9] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
3034 				p->idx += 10;
3035 				break;
3036 			/* Copy L2T/T2L, partial */
3037 			case 0x49:
3038 				/* L2T, T2L partial */
3039 				if (p->family < CHIP_CAYMAN) {
3040 					DRM_ERROR("L2T, T2L Partial is cayman only !\n");
3041 					return -EINVAL;
3042 				}
3043 				/* detile bit */
3044 				if (radeon_get_ib_value(p, idx + 2) & (1 << 31)) {
3045 					/* tiled src, linear dst */
3046 					ib[idx+1] += (u32)(src_reloc->gpu_offset >> 8);
3047 
3048 					ib[idx+7] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
3049 					ib[idx+8] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
3050 				} else {
3051 					/* linear src, tiled dst */
3052 					ib[idx+7] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
3053 					ib[idx+8] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
3054 
3055 					ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
3056 				}
3057 				p->idx += 12;
3058 				break;
3059 			/* Copy L2T broadcast */
3060 			case 0x4b:
3061 				/* L2T, broadcast */
3062 				if (radeon_get_ib_value(p, idx + 2) & (1 << 31)) {
3063 					DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
3064 					return -EINVAL;
3065 				}
3066 				r = r600_dma_cs_next_reloc(p, &dst2_reloc);
3067 				if (r) {
3068 					DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
3069 					return -EINVAL;
3070 				}
3071 				dst_offset = radeon_get_ib_value(p, idx+1);
3072 				dst_offset <<= 8;
3073 				dst2_offset = radeon_get_ib_value(p, idx+2);
3074 				dst2_offset <<= 8;
3075 				src_offset = radeon_get_ib_value(p, idx+8);
3076 				src_offset |= ((u64)(radeon_get_ib_value(p, idx+9) & 0xff)) << 32;
3077 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3078 					dev_warn(p->dev, "DMA L2T, broadcast src buffer too small (%llu %lu)\n",
3079 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3080 					return -EINVAL;
3081 				}
3082 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3083 					dev_warn(p->dev, "DMA L2T, broadcast dst buffer too small (%llu %lu)\n",
3084 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3085 					return -EINVAL;
3086 				}
3087 				if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
3088 					dev_warn(p->dev, "DMA L2T, broadcast dst2 buffer too small (%llu %lu)\n",
3089 							dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
3090 					return -EINVAL;
3091 				}
3092 				ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
3093 				ib[idx+2] += (u32)(dst2_reloc->gpu_offset >> 8);
3094 				ib[idx+8] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
3095 				ib[idx+9] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
3096 				p->idx += 10;
3097 				break;
3098 			/* Copy L2T/T2L (tile units) */
3099 			case 0x4c:
3100 				/* L2T, T2L */
3101 				/* detile bit */
3102 				if (radeon_get_ib_value(p, idx + 2) & (1 << 31)) {
3103 					/* tiled src, linear dst */
3104 					src_offset = radeon_get_ib_value(p, idx+1);
3105 					src_offset <<= 8;
3106 					ib[idx+1] += (u32)(src_reloc->gpu_offset >> 8);
3107 
3108 					dst_offset = radeon_get_ib_value(p, idx+7);
3109 					dst_offset |= ((u64)(radeon_get_ib_value(p, idx+8) & 0xff)) << 32;
3110 					ib[idx+7] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
3111 					ib[idx+8] += upper_32_bits(dst_reloc->gpu_offset) & 0xff;
3112 				} else {
3113 					/* linear src, tiled dst */
3114 					src_offset = radeon_get_ib_value(p, idx+7);
3115 					src_offset |= ((u64)(radeon_get_ib_value(p, idx+8) & 0xff)) << 32;
3116 					ib[idx+7] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
3117 					ib[idx+8] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
3118 
3119 					dst_offset = radeon_get_ib_value(p, idx+1);
3120 					dst_offset <<= 8;
3121 					ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
3122 				}
3123 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3124 					dev_warn(p->dev, "DMA L2T, T2L src buffer too small (%llu %lu)\n",
3125 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3126 					return -EINVAL;
3127 				}
3128 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3129 					dev_warn(p->dev, "DMA L2T, T2L dst buffer too small (%llu %lu)\n",
3130 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3131 					return -EINVAL;
3132 				}
3133 				p->idx += 9;
3134 				break;
3135 			/* Copy T2T, partial (tile units) */
3136 			case 0x4d:
3137 				/* T2T partial */
3138 				if (p->family < CHIP_CAYMAN) {
3139 					DRM_ERROR("L2T, T2L Partial is cayman only !\n");
3140 					return -EINVAL;
3141 				}
3142 				ib[idx+1] += (u32)(src_reloc->gpu_offset >> 8);
3143 				ib[idx+4] += (u32)(dst_reloc->gpu_offset >> 8);
3144 				p->idx += 13;
3145 				break;
3146 			/* Copy L2T broadcast (tile units) */
3147 			case 0x4f:
3148 				/* L2T, broadcast */
3149 				if (radeon_get_ib_value(p, idx + 2) & (1 << 31)) {
3150 					DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
3151 					return -EINVAL;
3152 				}
3153 				r = r600_dma_cs_next_reloc(p, &dst2_reloc);
3154 				if (r) {
3155 					DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
3156 					return -EINVAL;
3157 				}
3158 				dst_offset = radeon_get_ib_value(p, idx+1);
3159 				dst_offset <<= 8;
3160 				dst2_offset = radeon_get_ib_value(p, idx+2);
3161 				dst2_offset <<= 8;
3162 				src_offset = radeon_get_ib_value(p, idx+8);
3163 				src_offset |= ((u64)(radeon_get_ib_value(p, idx+9) & 0xff)) << 32;
3164 				if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
3165 					dev_warn(p->dev, "DMA L2T, broadcast src buffer too small (%llu %lu)\n",
3166 							src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
3167 					return -EINVAL;
3168 				}
3169 				if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3170 					dev_warn(p->dev, "DMA L2T, broadcast dst buffer too small (%llu %lu)\n",
3171 							dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
3172 					return -EINVAL;
3173 				}
3174 				if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
3175 					dev_warn(p->dev, "DMA L2T, broadcast dst2 buffer too small (%llu %lu)\n",
3176 							dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
3177 					return -EINVAL;
3178 				}
3179 				ib[idx+1] += (u32)(dst_reloc->gpu_offset >> 8);
3180 				ib[idx+2] += (u32)(dst2_reloc->gpu_offset >> 8);
3181 				ib[idx+8] += (u32)(src_reloc->gpu_offset & 0xfffffffc);
3182 				ib[idx+9] += upper_32_bits(src_reloc->gpu_offset) & 0xff;
3183 				p->idx += 10;
3184 				break;
3185 			default:
3186 				DRM_ERROR("bad DMA_PACKET_COPY [%6d] 0x%08x invalid sub cmd\n", idx, header);
3187 				return -EINVAL;
3188 			}
3189 			break;
3190 		case DMA_PACKET_CONSTANT_FILL:
3191 			r = r600_dma_cs_next_reloc(p, &dst_reloc);
3192 			if (r) {
3193 				DRM_ERROR("bad DMA_PACKET_CONSTANT_FILL\n");
3194 				return -EINVAL;
3195 			}
3196 			dst_offset = radeon_get_ib_value(p, idx+1);
3197 			dst_offset |= ((u64)(radeon_get_ib_value(p, idx+3) & 0x00ff0000)) << 16;
3198 			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
3199 				dev_warn(p->dev, "DMA constant fill buffer too small (%llu %lu)\n",
3200 					 dst_offset, radeon_bo_size(dst_reloc->robj));
3201 				return -EINVAL;
3202 			}
3203 			ib[idx+1] += (u32)(dst_reloc->gpu_offset & 0xfffffffc);
3204 			ib[idx+3] += (upper_32_bits(dst_reloc->gpu_offset) << 16) & 0x00ff0000;
3205 			p->idx += 4;
3206 			break;
3207 		case DMA_PACKET_NOP:
3208 			p->idx += 1;
3209 			break;
3210 		default:
3211 			DRM_ERROR("Unknown packet type %d at %d !\n", cmd, idx);
3212 			return -EINVAL;
3213 		}
3214 	} while (p->idx < p->chunk_ib->length_dw);
3215 #if 0
3216 	for (r = 0; r < p->ib->length_dw; r++) {
3217 		pr_info("%05d  0x%08X\n", r, p->ib.ptr[r]);
3218 		mdelay(1);
3219 	}
3220 #endif
3221 	return 0;
3222 }
3223 
3224 /* vm parser */
3225 static bool evergreen_vm_reg_valid(u32 reg)
3226 {
3227 	/* context regs are fine */
3228 	if (reg >= 0x28000)
3229 		return true;
3230 
3231 	/* check config regs */
3232 	switch (reg) {
3233 	case WAIT_UNTIL:
3234 	case GRBM_GFX_INDEX:
3235 	case CP_STRMOUT_CNTL:
3236 	case CP_COHER_CNTL:
3237 	case CP_COHER_SIZE:
3238 	case VGT_VTX_VECT_EJECT_REG:
3239 	case VGT_CACHE_INVALIDATION:
3240 	case VGT_GS_VERTEX_REUSE:
3241 	case VGT_PRIMITIVE_TYPE:
3242 	case VGT_INDEX_TYPE:
3243 	case VGT_NUM_INDICES:
3244 	case VGT_NUM_INSTANCES:
3245 	case VGT_COMPUTE_DIM_X:
3246 	case VGT_COMPUTE_DIM_Y:
3247 	case VGT_COMPUTE_DIM_Z:
3248 	case VGT_COMPUTE_START_X:
3249 	case VGT_COMPUTE_START_Y:
3250 	case VGT_COMPUTE_START_Z:
3251 	case VGT_COMPUTE_INDEX:
3252 	case VGT_COMPUTE_THREAD_GROUP_SIZE:
3253 	case VGT_HS_OFFCHIP_PARAM:
3254 	case PA_CL_ENHANCE:
3255 	case PA_SU_LINE_STIPPLE_VALUE:
3256 	case PA_SC_LINE_STIPPLE_STATE:
3257 	case PA_SC_ENHANCE:
3258 	case SQ_DYN_GPR_CNTL_PS_FLUSH_REQ:
3259 	case SQ_DYN_GPR_SIMD_LOCK_EN:
3260 	case SQ_CONFIG:
3261 	case SQ_GPR_RESOURCE_MGMT_1:
3262 	case SQ_GLOBAL_GPR_RESOURCE_MGMT_1:
3263 	case SQ_GLOBAL_GPR_RESOURCE_MGMT_2:
3264 	case SQ_CONST_MEM_BASE:
3265 	case SQ_STATIC_THREAD_MGMT_1:
3266 	case SQ_STATIC_THREAD_MGMT_2:
3267 	case SQ_STATIC_THREAD_MGMT_3:
3268 	case SPI_CONFIG_CNTL:
3269 	case SPI_CONFIG_CNTL_1:
3270 	case TA_CNTL_AUX:
3271 	case DB_DEBUG:
3272 	case DB_DEBUG2:
3273 	case DB_DEBUG3:
3274 	case DB_DEBUG4:
3275 	case DB_WATERMARKS:
3276 	case TD_PS_BORDER_COLOR_INDEX:
3277 	case TD_PS_BORDER_COLOR_RED:
3278 	case TD_PS_BORDER_COLOR_GREEN:
3279 	case TD_PS_BORDER_COLOR_BLUE:
3280 	case TD_PS_BORDER_COLOR_ALPHA:
3281 	case TD_VS_BORDER_COLOR_INDEX:
3282 	case TD_VS_BORDER_COLOR_RED:
3283 	case TD_VS_BORDER_COLOR_GREEN:
3284 	case TD_VS_BORDER_COLOR_BLUE:
3285 	case TD_VS_BORDER_COLOR_ALPHA:
3286 	case TD_GS_BORDER_COLOR_INDEX:
3287 	case TD_GS_BORDER_COLOR_RED:
3288 	case TD_GS_BORDER_COLOR_GREEN:
3289 	case TD_GS_BORDER_COLOR_BLUE:
3290 	case TD_GS_BORDER_COLOR_ALPHA:
3291 	case TD_HS_BORDER_COLOR_INDEX:
3292 	case TD_HS_BORDER_COLOR_RED:
3293 	case TD_HS_BORDER_COLOR_GREEN:
3294 	case TD_HS_BORDER_COLOR_BLUE:
3295 	case TD_HS_BORDER_COLOR_ALPHA:
3296 	case TD_LS_BORDER_COLOR_INDEX:
3297 	case TD_LS_BORDER_COLOR_RED:
3298 	case TD_LS_BORDER_COLOR_GREEN:
3299 	case TD_LS_BORDER_COLOR_BLUE:
3300 	case TD_LS_BORDER_COLOR_ALPHA:
3301 	case TD_CS_BORDER_COLOR_INDEX:
3302 	case TD_CS_BORDER_COLOR_RED:
3303 	case TD_CS_BORDER_COLOR_GREEN:
3304 	case TD_CS_BORDER_COLOR_BLUE:
3305 	case TD_CS_BORDER_COLOR_ALPHA:
3306 	case SQ_ESGS_RING_SIZE:
3307 	case SQ_GSVS_RING_SIZE:
3308 	case SQ_ESTMP_RING_SIZE:
3309 	case SQ_GSTMP_RING_SIZE:
3310 	case SQ_HSTMP_RING_SIZE:
3311 	case SQ_LSTMP_RING_SIZE:
3312 	case SQ_PSTMP_RING_SIZE:
3313 	case SQ_VSTMP_RING_SIZE:
3314 	case SQ_ESGS_RING_ITEMSIZE:
3315 	case SQ_ESTMP_RING_ITEMSIZE:
3316 	case SQ_GSTMP_RING_ITEMSIZE:
3317 	case SQ_GSVS_RING_ITEMSIZE:
3318 	case SQ_GS_VERT_ITEMSIZE:
3319 	case SQ_GS_VERT_ITEMSIZE_1:
3320 	case SQ_GS_VERT_ITEMSIZE_2:
3321 	case SQ_GS_VERT_ITEMSIZE_3:
3322 	case SQ_GSVS_RING_OFFSET_1:
3323 	case SQ_GSVS_RING_OFFSET_2:
3324 	case SQ_GSVS_RING_OFFSET_3:
3325 	case SQ_HSTMP_RING_ITEMSIZE:
3326 	case SQ_LSTMP_RING_ITEMSIZE:
3327 	case SQ_PSTMP_RING_ITEMSIZE:
3328 	case SQ_VSTMP_RING_ITEMSIZE:
3329 	case VGT_TF_RING_SIZE:
3330 	case SQ_ESGS_RING_BASE:
3331 	case SQ_GSVS_RING_BASE:
3332 	case SQ_ESTMP_RING_BASE:
3333 	case SQ_GSTMP_RING_BASE:
3334 	case SQ_HSTMP_RING_BASE:
3335 	case SQ_LSTMP_RING_BASE:
3336 	case SQ_PSTMP_RING_BASE:
3337 	case SQ_VSTMP_RING_BASE:
3338 	case CAYMAN_VGT_OFFCHIP_LDS_BASE:
3339 	case CAYMAN_SQ_EX_ALLOC_TABLE_SLOTS:
3340 		return true;
3341 	default:
3342 		DRM_ERROR("Invalid register 0x%x in CS\n", reg);
3343 		return false;
3344 	}
3345 }
3346 
3347 static int evergreen_vm_packet3_check(struct radeon_device *rdev,
3348 				      u32 *ib, struct radeon_cs_packet *pkt)
3349 {
3350 	u32 idx = pkt->idx + 1;
3351 	u32 idx_value = ib[idx];
3352 	u32 start_reg, end_reg, reg, i;
3353 	u32 command, info;
3354 
3355 	switch (pkt->opcode) {
3356 	case PACKET3_NOP:
3357 		break;
3358 	case PACKET3_SET_BASE:
3359 		if (idx_value != 1) {
3360 			DRM_ERROR("bad SET_BASE");
3361 			return -EINVAL;
3362 		}
3363 		break;
3364 	case PACKET3_CLEAR_STATE:
3365 	case PACKET3_INDEX_BUFFER_SIZE:
3366 	case PACKET3_DISPATCH_DIRECT:
3367 	case PACKET3_DISPATCH_INDIRECT:
3368 	case PACKET3_MODE_CONTROL:
3369 	case PACKET3_SET_PREDICATION:
3370 	case PACKET3_COND_EXEC:
3371 	case PACKET3_PRED_EXEC:
3372 	case PACKET3_DRAW_INDIRECT:
3373 	case PACKET3_DRAW_INDEX_INDIRECT:
3374 	case PACKET3_INDEX_BASE:
3375 	case PACKET3_DRAW_INDEX_2:
3376 	case PACKET3_CONTEXT_CONTROL:
3377 	case PACKET3_DRAW_INDEX_OFFSET:
3378 	case PACKET3_INDEX_TYPE:
3379 	case PACKET3_DRAW_INDEX:
3380 	case PACKET3_DRAW_INDEX_AUTO:
3381 	case PACKET3_DRAW_INDEX_IMMD:
3382 	case PACKET3_NUM_INSTANCES:
3383 	case PACKET3_DRAW_INDEX_MULTI_AUTO:
3384 	case PACKET3_STRMOUT_BUFFER_UPDATE:
3385 	case PACKET3_DRAW_INDEX_OFFSET_2:
3386 	case PACKET3_DRAW_INDEX_MULTI_ELEMENT:
3387 	case PACKET3_MPEG_INDEX:
3388 	case PACKET3_WAIT_REG_MEM:
3389 	case PACKET3_MEM_WRITE:
3390 	case PACKET3_PFP_SYNC_ME:
3391 	case PACKET3_SURFACE_SYNC:
3392 	case PACKET3_EVENT_WRITE:
3393 	case PACKET3_EVENT_WRITE_EOP:
3394 	case PACKET3_EVENT_WRITE_EOS:
3395 	case PACKET3_SET_CONTEXT_REG:
3396 	case PACKET3_SET_BOOL_CONST:
3397 	case PACKET3_SET_LOOP_CONST:
3398 	case PACKET3_SET_RESOURCE:
3399 	case PACKET3_SET_SAMPLER:
3400 	case PACKET3_SET_CTL_CONST:
3401 	case PACKET3_SET_RESOURCE_OFFSET:
3402 	case PACKET3_SET_CONTEXT_REG_INDIRECT:
3403 	case PACKET3_SET_RESOURCE_INDIRECT:
3404 	case CAYMAN_PACKET3_DEALLOC_STATE:
3405 		break;
3406 	case PACKET3_COND_WRITE:
3407 		if (idx_value & 0x100) {
3408 			reg = ib[idx + 5] * 4;
3409 			if (!evergreen_vm_reg_valid(reg))
3410 				return -EINVAL;
3411 		}
3412 		break;
3413 	case PACKET3_COPY_DW:
3414 		if (idx_value & 0x2) {
3415 			reg = ib[idx + 3] * 4;
3416 			if (!evergreen_vm_reg_valid(reg))
3417 				return -EINVAL;
3418 		}
3419 		break;
3420 	case PACKET3_SET_CONFIG_REG:
3421 		start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START;
3422 		end_reg = 4 * pkt->count + start_reg - 4;
3423 		if ((start_reg < PACKET3_SET_CONFIG_REG_START) ||
3424 		    (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
3425 		    (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
3426 			DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
3427 			return -EINVAL;
3428 		}
3429 		for (i = 0; i < pkt->count; i++) {
3430 			reg = start_reg + (4 * i);
3431 			if (!evergreen_vm_reg_valid(reg))
3432 				return -EINVAL;
3433 		}
3434 		break;
3435 	case PACKET3_CP_DMA:
3436 		command = ib[idx + 4];
3437 		info = ib[idx + 1];
3438 		if ((((info & 0x60000000) >> 29) != 0) || /* src = GDS or DATA */
3439 		    (((info & 0x00300000) >> 20) != 0) || /* dst = GDS */
3440 		    ((((info & 0x00300000) >> 20) == 0) &&
3441 		     (command & PACKET3_CP_DMA_CMD_DAS)) || /* dst = register */
3442 		    ((((info & 0x60000000) >> 29) == 0) &&
3443 		     (command & PACKET3_CP_DMA_CMD_SAS))) { /* src = register */
3444 			/* non mem to mem copies requires dw aligned count */
3445 			if ((command & 0x1fffff) % 4) {
3446 				DRM_ERROR("CP DMA command requires dw count alignment\n");
3447 				return -EINVAL;
3448 			}
3449 		}
3450 		if (command & PACKET3_CP_DMA_CMD_SAS) {
3451 			/* src address space is register */
3452 			if (((info & 0x60000000) >> 29) == 0) {
3453 				start_reg = idx_value << 2;
3454 				if (command & PACKET3_CP_DMA_CMD_SAIC) {
3455 					reg = start_reg;
3456 					if (!evergreen_vm_reg_valid(reg)) {
3457 						DRM_ERROR("CP DMA Bad SRC register\n");
3458 						return -EINVAL;
3459 					}
3460 				} else {
3461 					for (i = 0; i < (command & 0x1fffff); i++) {
3462 						reg = start_reg + (4 * i);
3463 						if (!evergreen_vm_reg_valid(reg)) {
3464 							DRM_ERROR("CP DMA Bad SRC register\n");
3465 							return -EINVAL;
3466 						}
3467 					}
3468 				}
3469 			}
3470 		}
3471 		if (command & PACKET3_CP_DMA_CMD_DAS) {
3472 			/* dst address space is register */
3473 			if (((info & 0x00300000) >> 20) == 0) {
3474 				start_reg = ib[idx + 2];
3475 				if (command & PACKET3_CP_DMA_CMD_DAIC) {
3476 					reg = start_reg;
3477 					if (!evergreen_vm_reg_valid(reg)) {
3478 						DRM_ERROR("CP DMA Bad DST register\n");
3479 						return -EINVAL;
3480 					}
3481 				} else {
3482 					for (i = 0; i < (command & 0x1fffff); i++) {
3483 						reg = start_reg + (4 * i);
3484 						if (!evergreen_vm_reg_valid(reg)) {
3485 							DRM_ERROR("CP DMA Bad DST register\n");
3486 							return -EINVAL;
3487 						}
3488 					}
3489 				}
3490 			}
3491 		}
3492 		break;
3493 	case PACKET3_SET_APPEND_CNT: {
3494 		uint32_t areg;
3495 		uint32_t allowed_reg_base;
3496 
3497 		if (pkt->count != 2) {
3498 			DRM_ERROR("bad SET_APPEND_CNT (invalid count)\n");
3499 			return -EINVAL;
3500 		}
3501 
3502 		allowed_reg_base = GDS_APPEND_COUNT_0;
3503 		allowed_reg_base -= PACKET3_SET_CONTEXT_REG_START;
3504 		allowed_reg_base >>= 2;
3505 
3506 		areg = idx_value >> 16;
3507 		if (areg < allowed_reg_base || areg > (allowed_reg_base + 11)) {
3508 			DRM_ERROR("forbidden register for append cnt 0x%08x at %d\n",
3509 				  areg, idx);
3510 			return -EINVAL;
3511 		}
3512 		break;
3513 	}
3514 	default:
3515 		return -EINVAL;
3516 	}
3517 	return 0;
3518 }
3519 
3520 int evergreen_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
3521 {
3522 	int ret = 0;
3523 	u32 idx = 0;
3524 	struct radeon_cs_packet pkt;
3525 
3526 	do {
3527 		pkt.idx = idx;
3528 		pkt.type = RADEON_CP_PACKET_GET_TYPE(ib->ptr[idx]);
3529 		pkt.count = RADEON_CP_PACKET_GET_COUNT(ib->ptr[idx]);
3530 		pkt.one_reg_wr = 0;
3531 		switch (pkt.type) {
3532 		case RADEON_PACKET_TYPE0:
3533 			dev_err(rdev->dev, "Packet0 not allowed!\n");
3534 			ret = -EINVAL;
3535 			break;
3536 		case RADEON_PACKET_TYPE2:
3537 			idx += 1;
3538 			break;
3539 		case RADEON_PACKET_TYPE3:
3540 			pkt.opcode = RADEON_CP_PACKET3_GET_OPCODE(ib->ptr[idx]);
3541 			ret = evergreen_vm_packet3_check(rdev, ib->ptr, &pkt);
3542 			idx += pkt.count + 2;
3543 			break;
3544 		default:
3545 			dev_err(rdev->dev, "Unknown packet type %d !\n", pkt.type);
3546 			ret = -EINVAL;
3547 			break;
3548 		}
3549 		if (ret)
3550 			break;
3551 	} while (idx < ib->length_dw);
3552 
3553 	return ret;
3554 }
3555 
3556 /**
3557  * evergreen_dma_ib_parse() - parse the DMA IB for VM
3558  * @rdev: radeon_device pointer
3559  * @ib:	radeon_ib pointer
3560  *
3561  * Parses the DMA IB from the VM CS ioctl
3562  * checks for errors. (Cayman-SI)
3563  * Returns 0 for success and an error on failure.
3564  **/
3565 int evergreen_dma_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
3566 {
3567 	u32 idx = 0;
3568 	u32 header, cmd, count, sub_cmd;
3569 
3570 	do {
3571 		header = ib->ptr[idx];
3572 		cmd = GET_DMA_CMD(header);
3573 		count = GET_DMA_COUNT(header);
3574 		sub_cmd = GET_DMA_SUB_CMD(header);
3575 
3576 		switch (cmd) {
3577 		case DMA_PACKET_WRITE:
3578 			switch (sub_cmd) {
3579 			/* tiled */
3580 			case 8:
3581 				idx += count + 7;
3582 				break;
3583 			/* linear */
3584 			case 0:
3585 				idx += count + 3;
3586 				break;
3587 			default:
3588 				DRM_ERROR("bad DMA_PACKET_WRITE [%6d] 0x%08x sub cmd is not 0 or 8\n", idx, ib->ptr[idx]);
3589 				return -EINVAL;
3590 			}
3591 			break;
3592 		case DMA_PACKET_COPY:
3593 			switch (sub_cmd) {
3594 			/* Copy L2L, DW aligned */
3595 			case 0x00:
3596 				idx += 5;
3597 				break;
3598 			/* Copy L2T/T2L */
3599 			case 0x08:
3600 				idx += 9;
3601 				break;
3602 			/* Copy L2L, byte aligned */
3603 			case 0x40:
3604 				idx += 5;
3605 				break;
3606 			/* Copy L2L, partial */
3607 			case 0x41:
3608 				idx += 9;
3609 				break;
3610 			/* Copy L2L, DW aligned, broadcast */
3611 			case 0x44:
3612 				idx += 7;
3613 				break;
3614 			/* Copy L2T Frame to Field */
3615 			case 0x48:
3616 				idx += 10;
3617 				break;
3618 			/* Copy L2T/T2L, partial */
3619 			case 0x49:
3620 				idx += 12;
3621 				break;
3622 			/* Copy L2T broadcast */
3623 			case 0x4b:
3624 				idx += 10;
3625 				break;
3626 			/* Copy L2T/T2L (tile units) */
3627 			case 0x4c:
3628 				idx += 9;
3629 				break;
3630 			/* Copy T2T, partial (tile units) */
3631 			case 0x4d:
3632 				idx += 13;
3633 				break;
3634 			/* Copy L2T broadcast (tile units) */
3635 			case 0x4f:
3636 				idx += 10;
3637 				break;
3638 			default:
3639 				DRM_ERROR("bad DMA_PACKET_COPY [%6d] 0x%08x invalid sub cmd\n", idx, ib->ptr[idx]);
3640 				return -EINVAL;
3641 			}
3642 			break;
3643 		case DMA_PACKET_CONSTANT_FILL:
3644 			idx += 4;
3645 			break;
3646 		case DMA_PACKET_NOP:
3647 			idx += 1;
3648 			break;
3649 		default:
3650 			DRM_ERROR("Unknown packet type %d at %d !\n", cmd, idx);
3651 			return -EINVAL;
3652 		}
3653 	} while (idx < ib->length_dw);
3654 
3655 	return 0;
3656 }
3657